[mobile] Patch faces mvp from photos-app repo

mobile/android/app/build.gradle

@@ -47,7 +47,7 @@ android {
 
     defaultConfig {
         applicationId "io.ente.photos"
-        minSdkVersion 21
+        minSdkVersion 26
         targetSdkVersion 33
         versionCode flutterVersionCode.toInteger()
         versionName flutterVersionName
@@ -74,6 +74,10 @@ android {
             dimension "default"
             applicationIdSuffix ".dev"
         }
+        face {
+            dimension "default"
+            applicationIdSuffix ".face"
+        }
         playstore {
             dimension "default"
         }

mobile/android/app/src/face/AndroidManifest.xml

@@ -0,0 +1,10 @@
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="io.ente.photos">
+    <!-- Flutter needs it to communicate with the running application
+         to allow setting breakpoints, to provide hot reload, etc.
+    -->
+    <uses-permission android:name="android.permission.INTERNET"/>
+    <uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE" />
+    <uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE" />
+    <uses-permission android:name="android.permission.ACCESS_NETWORK_STATE" />
+</manifest>

mobile/android/app/src/face/res/values/strings.xml

@@ -0,0 +1,4 @@
+<resources>
+    <string name="app_name">ente face</string>
+    <string name="backup">backup face</string>
+</resources>

mobile/ios/Podfile.lock

@@ -59,6 +59,8 @@ PODS:
   - flutter_inappwebview/Core (0.0.1):
     - Flutter
     - OrderedSet (~> 5.0)
+  - flutter_isolate (0.0.1):
+    - Flutter
   - flutter_local_notifications (0.0.1):
     - Flutter
   - flutter_native_splash (0.0.1):
@@ -197,6 +199,28 @@ PODS:
     - sqlite3/fts5
     - sqlite3/perf-threadsafe
     - sqlite3/rtree
+  - TensorFlowLiteC (2.12.0):
+    - TensorFlowLiteC/Core (= 2.12.0)
+  - TensorFlowLiteC/Core (2.12.0)
+  - TensorFlowLiteC/CoreML (2.12.0):
+    - TensorFlowLiteC/Core
+  - TensorFlowLiteC/Metal (2.12.0):
+    - TensorFlowLiteC/Core
+  - TensorFlowLiteSwift (2.12.0):
+    - TensorFlowLiteSwift/Core (= 2.12.0)
+  - TensorFlowLiteSwift/Core (2.12.0):
+    - TensorFlowLiteC (= 2.12.0)
+  - TensorFlowLiteSwift/CoreML (2.12.0):
+    - TensorFlowLiteC/CoreML (= 2.12.0)
+    - TensorFlowLiteSwift/Core (= 2.12.0)
+  - TensorFlowLiteSwift/Metal (2.12.0):
+    - TensorFlowLiteC/Metal (= 2.12.0)
+    - TensorFlowLiteSwift/Core (= 2.12.0)
+  - tflite_flutter (0.0.1):
+    - Flutter
+    - TensorFlowLiteSwift (= 2.12.0)
+    - TensorFlowLiteSwift/CoreML (= 2.12.0)
+    - TensorFlowLiteSwift/Metal (= 2.12.0)
   - Toast (4.1.0)
   - uni_links (0.0.1):
     - Flutter
@@ -228,6 +252,7 @@ DEPENDENCIES:
   - flutter_email_sender (from `.symlinks/plugins/flutter_email_sender/ios`)
   - flutter_image_compress (from `.symlinks/plugins/flutter_image_compress/ios`)
   - flutter_inappwebview (from `.symlinks/plugins/flutter_inappwebview/ios`)
+  - flutter_isolate (from `.symlinks/plugins/flutter_isolate/ios`)
   - flutter_local_notifications (from `.symlinks/plugins/flutter_local_notifications/ios`)
   - flutter_native_splash (from `.symlinks/plugins/flutter_native_splash/ios`)
   - flutter_secure_storage (from `.symlinks/plugins/flutter_secure_storage/ios`)
@@ -257,6 +282,7 @@ DEPENDENCIES:
   - shared_preferences_foundation (from `.symlinks/plugins/shared_preferences_foundation/darwin`)
   - sqflite (from `.symlinks/plugins/sqflite/darwin`)
   - sqlite3_flutter_libs (from `.symlinks/plugins/sqlite3_flutter_libs/ios`)
+  - tflite_flutter (from `.symlinks/plugins/tflite_flutter/ios`)
   - uni_links (from `.symlinks/plugins/uni_links/ios`)
   - url_launcher_ios (from `.symlinks/plugins/url_launcher_ios/ios`)
   - video_player_avfoundation (from `.symlinks/plugins/video_player_avfoundation/darwin`)
@@ -287,6 +313,8 @@ SPEC REPOS:
     - Sentry
     - SentryPrivate
     - sqlite3
+    - TensorFlowLiteC
+    - TensorFlowLiteSwift
     - Toast
 
 EXTERNAL SOURCES:
@@ -314,6 +342,8 @@ EXTERNAL SOURCES:
     :path: ".symlinks/plugins/flutter_image_compress/ios"
   flutter_inappwebview:
     :path: ".symlinks/plugins/flutter_inappwebview/ios"
+  flutter_isolate:
+    :path: ".symlinks/plugins/flutter_isolate/ios"
   flutter_local_notifications:
     :path: ".symlinks/plugins/flutter_local_notifications/ios"
   flutter_native_splash:
@@ -372,6 +402,8 @@ EXTERNAL SOURCES:
     :path: ".symlinks/plugins/sqflite/darwin"
   sqlite3_flutter_libs:
     :path: ".symlinks/plugins/sqlite3_flutter_libs/ios"
+  tflite_flutter:
+    :path: ".symlinks/plugins/tflite_flutter/ios"
   uni_links:
     :path: ".symlinks/plugins/uni_links/ios"
   url_launcher_ios:
@@ -405,6 +437,7 @@ SPEC CHECKSUMS:
   flutter_email_sender: 02d7443217d8c41483223627972bfdc09f74276b
   flutter_image_compress: 5a5e9aee05b6553048b8df1c3bc456d0afaac433
   flutter_inappwebview: 3d32228f1304635e7c028b0d4252937730bbc6cf
+  flutter_isolate: 0edf5081826d071adf21759d1eb10ff5c24503b5
   flutter_local_notifications: 0c0b1ae97e741e1521e4c1629a459d04b9aec743
   flutter_native_splash: 52501b97d1c0a5f898d687f1646226c1f93c56ef
   flutter_secure_storage: 23fc622d89d073675f2eaa109381aefbcf5a49be
@@ -449,6 +482,9 @@ SPEC CHECKSUMS:
   sqflite: 673a0e54cc04b7d6dba8d24fb8095b31c3a99eec
   sqlite3: 73b7fc691fdc43277614250e04d183740cb15078
   sqlite3_flutter_libs: aeb4d37509853dfa79d9b59386a2dac5dd079428
+  TensorFlowLiteC: 20785a69299185a379ba9852b6625f00afd7984a
+  TensorFlowLiteSwift: 3a4928286e9e35bdd3e17970f48e53c80d25e793
+  tflite_flutter: 9433d086a3060431bbc9f3c7c20d017db0e72d08
   Toast: ec33c32b8688982cecc6348adeae667c1b9938da
   uni_links: d97da20c7701486ba192624d99bffaaffcfc298a
   url_launcher_ios: bbd758c6e7f9fd7b5b1d4cde34d2b95fcce5e812

mobile/ios/Runner.xcodeproj/project.pbxproj

@@ -299,6 +299,7 @@
 				"${BUILT_PRODUCTS_DIR}/flutter_email_sender/flutter_email_sender.framework",
 				"${BUILT_PRODUCTS_DIR}/flutter_image_compress/flutter_image_compress.framework",
 				"${BUILT_PRODUCTS_DIR}/flutter_inappwebview/flutter_inappwebview.framework",
+				"${BUILT_PRODUCTS_DIR}/flutter_isolate/flutter_isolate.framework",
 				"${BUILT_PRODUCTS_DIR}/flutter_local_notifications/flutter_local_notifications.framework",
 				"${BUILT_PRODUCTS_DIR}/flutter_native_splash/flutter_native_splash.framework",
 				"${BUILT_PRODUCTS_DIR}/flutter_secure_storage/flutter_secure_storage.framework",
@@ -382,6 +383,7 @@
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_email_sender.framework",
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_image_compress.framework",
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_inappwebview.framework",
+				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_isolate.framework",
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_local_notifications.framework",
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_native_splash.framework",
 				"${TARGET_BUILD_DIR}/${FRAMEWORKS_FOLDER_PATH}/flutter_secure_storage.framework",

mobile/lib/core/configuration.dart

@@ -18,6 +18,7 @@ import 'package:photos/db/trash_db.dart';
 import 'package:photos/db/upload_locks_db.dart';
 import 'package:photos/events/signed_in_event.dart';
 import 'package:photos/events/user_logged_out_event.dart';
+import "package:photos/face/db.dart";
 import 'package:photos/models/key_attributes.dart';
 import 'package:photos/models/key_gen_result.dart';
 import 'package:photos/models/private_key_attributes.dart';
@@ -164,6 +165,7 @@ class Configuration {
         : null;
     await CollectionsDB.instance.clearTable();
     await MemoriesDB.instance.clearTable();
+    await FaceMLDataDB.instance.clearTable();
 
     await UploadLocksDB.instance.clearTable();
     await IgnoredFilesService.instance.reset();

mobile/lib/db/ml_data_db.dart

@@ -0,0 +1,714 @@
+import 'dart:async';
+
+import 'package:logging/logging.dart';
+import 'package:path/path.dart' show join;
+import 'package:path_provider/path_provider.dart';
+import 'package:photos/models/ml/ml_typedefs.dart';
+import "package:photos/services/face_ml/face_feedback.dart/cluster_feedback.dart";
+import "package:photos/services/face_ml/face_feedback.dart/feedback_types.dart";
+import "package:photos/services/face_ml/face_ml_result.dart";
+import 'package:sqflite/sqflite.dart';
+
+/// Stores all data for the ML-related features. The database can be accessed by `MlDataDB.instance.database`.
+///
+/// This includes:
+/// [facesTable] - Stores all the detected faces and its embeddings in the images.
+/// [peopleTable] - Stores all the clusters of faces which are considered to be the same person.
+class MlDataDB {
+  static final Logger _logger = Logger("MlDataDB");
+
+  // TODO: [BOB] put the db in files
+  static const _databaseName = "ente.ml_data.db";
+  static const _databaseVersion = 1;
+
+  static const facesTable = 'faces';
+  static const fileIDColumn = 'file_id';
+  static const faceMlResultColumn = 'face_ml_result';
+  static const mlVersionColumn = 'ml_version';
+
+  static const peopleTable = 'people';
+  static const personIDColumn = 'person_id';
+  static const clusterResultColumn = 'cluster_result';
+  static const centroidColumn = 'cluster_centroid';
+  static const centroidDistanceThresholdColumn = 'centroid_distance_threshold';
+
+  static const feedbackTable = 'feedback';
+  static const feedbackIDColumn = 'feedback_id';
+  static const feedbackTypeColumn = 'feedback_type';
+  static const feedbackDataColumn = 'feedback_data';
+  static const feedbackTimestampColumn = 'feedback_timestamp';
+  static const feedbackFaceMlVersionColumn = 'feedback_face_ml_version';
+  static const feedbackClusterMlVersionColumn = 'feedback_cluster_ml_version';
+
+  static const createFacesTable = '''CREATE TABLE IF NOT EXISTS $facesTable (
+  $fileIDColumn	INTEGER NOT NULL UNIQUE,
+	$faceMlResultColumn	TEXT NOT NULL,
+  $mlVersionColumn	INTEGER NOT NULL,
+  PRIMARY KEY($fileIDColumn)
+  );
+  ''';
+  static const createPeopleTable = '''CREATE TABLE IF NOT EXISTS $peopleTable (
+  $personIDColumn	INTEGER NOT NULL UNIQUE,
+	$clusterResultColumn	TEXT NOT NULL,
+  $centroidColumn	TEXT NOT NULL,
+  $centroidDistanceThresholdColumn	REAL NOT NULL,
+	PRIMARY KEY($personIDColumn)
+  );
+  ''';
+  static const createFeedbackTable =
+      '''CREATE TABLE IF NOT EXISTS $feedbackTable (
+  $feedbackIDColumn	TEXT NOT NULL UNIQUE,
+  $feedbackTypeColumn	TEXT NOT NULL,
+  $feedbackDataColumn	TEXT NOT NULL,
+  $feedbackTimestampColumn	TEXT NOT NULL,
+  $feedbackFaceMlVersionColumn	INTEGER NOT NULL,
+  $feedbackClusterMlVersionColumn	INTEGER NOT NULL,
+  PRIMARY KEY($feedbackIDColumn)
+  );
+  ''';
+  static const _deleteFacesTable = 'DROP TABLE IF EXISTS $facesTable';
+  static const _deletePeopleTable = 'DROP TABLE IF EXISTS $peopleTable';
+  static const _deleteFeedbackTable = 'DROP TABLE IF EXISTS $feedbackTable';
+
+  MlDataDB._privateConstructor();
+  static final MlDataDB instance = MlDataDB._privateConstructor();
+
+  static Future<Database>? _dbFuture;
+  Future<Database> get database async {
+    _dbFuture ??= _initDatabase();
+    return _dbFuture!;
+  }
+
+  Future<Database> _initDatabase() async {
+    final documentsDirectory = await getApplicationDocumentsDirectory();
+    final String databaseDirectory =
+        join(documentsDirectory.path, _databaseName);
+    return await openDatabase(
+      databaseDirectory,
+      version: _databaseVersion,
+      onCreate: _onCreate,
+    );
+  }
+
+  Future _onCreate(Database db, int version) async {
+    await db.execute(createFacesTable);
+    await db.execute(createPeopleTable);
+    await db.execute(createFeedbackTable);
+  }
+
+  /// WARNING: This will delete ALL data in the database! Only use this for debug/testing purposes!
+  Future<void> cleanTables({
+    bool cleanFaces = false,
+    bool cleanPeople = false,
+    bool cleanFeedback = false,
+  }) async {
+    _logger.fine('`cleanTables()` called');
+    final db = await instance.database;
+
+    if (cleanFaces) {
+      _logger.fine('`cleanTables()`: Cleaning faces table');
+      await db.execute(_deleteFacesTable);
+    }
+
+    if (cleanPeople) {
+      _logger.fine('`cleanTables()`: Cleaning people table');
+      await db.execute(_deletePeopleTable);
+    }
+
+    if (cleanFeedback) {
+      _logger.fine('`cleanTables()`: Cleaning feedback table');
+      await db.execute(_deleteFeedbackTable);
+    }
+
+    if (!cleanFaces && !cleanPeople && !cleanFeedback) {
+      _logger.fine(
+        '`cleanTables()`: No tables cleaned, since no table was specified. Please be careful with this function!',
+      );
+    }
+
+    await db.execute(createFacesTable);
+    await db.execute(createPeopleTable);
+    await db.execute(createFeedbackTable);
+  }
+
+  Future<void> createFaceMlResult(FaceMlResult faceMlResult) async {
+    _logger.fine('createFaceMlResult called');
+
+    final existingResult = await getFaceMlResult(faceMlResult.fileId);
+    if (existingResult != null) {
+      if (faceMlResult.mlVersion <= existingResult.mlVersion) {
+        _logger.fine(
+          'FaceMlResult with file ID ${faceMlResult.fileId} already exists with equal or higher version. Skipping insert.',
+        );
+        return;
+      }
+    }
+
+    final db = await instance.database;
+    await db.insert(
+      facesTable,
+      {
+        fileIDColumn: faceMlResult.fileId,
+        faceMlResultColumn: faceMlResult.toJsonString(),
+        mlVersionColumn: faceMlResult.mlVersion,
+      },
+      conflictAlgorithm: ConflictAlgorithm.replace,
+    );
+  }
+
+  Future<bool> doesFaceMlResultExist(int fileId, {int? mlVersion}) async {
+    _logger.fine('doesFaceMlResultExist called');
+    final db = await instance.database;
+
+    String whereString = '$fileIDColumn = ?';
+    final List<dynamic> whereArgs = [fileId];
+
+    if (mlVersion != null) {
+      whereString += ' AND $mlVersionColumn = ?';
+      whereArgs.add(mlVersion);
+    }
+
+    final result = await db.query(
+      facesTable,
+      where: whereString,
+      whereArgs: whereArgs,
+      limit: 1,
+    );
+    return result.isNotEmpty;
+  }
+
+  Future<FaceMlResult?> getFaceMlResult(int fileId, {int? mlVersion}) async {
+    _logger.fine('getFaceMlResult called');
+    final db = await instance.database;
+
+    String whereString = '$fileIDColumn = ?';
+    final List<dynamic> whereArgs = [fileId];
+
+    if (mlVersion != null) {
+      whereString += ' AND $mlVersionColumn = ?';
+      whereArgs.add(mlVersion);
+    }
+
+    final result = await db.query(
+      facesTable,
+      where: whereString,
+      whereArgs: whereArgs,
+      limit: 1,
+    );
+    if (result.isNotEmpty) {
+      return FaceMlResult.fromJsonString(
+        result.first[faceMlResultColumn] as String,
+      );
+    }
+    _logger.fine(
+      'No faceMlResult found for fileID $fileId and mlVersion $mlVersion (null if not specified)',
+    );
+    return null;
+  }
+
+  /// Returns the faceMlResults for the given [fileIds].
+  Future<List<FaceMlResult>> getSelectedFaceMlResults(
+    List<int> fileIds,
+  ) async {
+    _logger.fine('getSelectedFaceMlResults called');
+    final db = await instance.database;
+
+    if (fileIds.isEmpty) {
+      _logger.warning('getSelectedFaceMlResults called with empty fileIds');
+      return <FaceMlResult>[];
+    }
+
+    final List<Map<String, Object?>> results = await db.query(
+      facesTable,
+      columns: [faceMlResultColumn],
+      where: '$fileIDColumn IN (${fileIds.join(',')})',
+      orderBy: fileIDColumn,
+    );
+
+    return results
+        .map(
+          (result) =>
+              FaceMlResult.fromJsonString(result[faceMlResultColumn] as String),
+        )
+        .toList();
+  }
+
+  Future<List<FaceMlResult>> getAllFaceMlResults({int? mlVersion}) async {
+    _logger.fine('getAllFaceMlResults called');
+    final db = await instance.database;
+
+    String? whereString;
+    List<dynamic>? whereArgs;
+
+    if (mlVersion != null) {
+      whereString = '$mlVersionColumn = ?';
+      whereArgs = [mlVersion];
+    }
+
+    final results = await db.query(
+      facesTable,
+      where: whereString,
+      whereArgs: whereArgs,
+      orderBy: fileIDColumn,
+    );
+
+    return results
+        .map(
+          (result) =>
+              FaceMlResult.fromJsonString(result[faceMlResultColumn] as String),
+        )
+        .toList();
+  }
+
+  /// getAllFileIDs returns a set of all fileIDs from the facesTable, meaning all the fileIDs for which a FaceMlResult exists, optionally filtered by mlVersion.
+  Future<Set<int>> getAllFaceMlResultFileIDs({int? mlVersion}) async {
+    _logger.fine('getAllFaceMlResultFileIDs called');
+    final db = await instance.database;
+
+    String? whereString;
+    List<dynamic>? whereArgs;
+
+    if (mlVersion != null) {
+      whereString = '$mlVersionColumn = ?';
+      whereArgs = [mlVersion];
+    }
+
+    final List<Map<String, Object?>> results = await db.query(
+      facesTable,
+      where: whereString,
+      whereArgs: whereArgs,
+      orderBy: fileIDColumn,
+    );
+
+    return results.map((result) => result[fileIDColumn] as int).toSet();
+  }
+
+  Future<Set<int>> getAllFaceMlResultFileIDsProcessedWithThumbnailOnly({
+    int? mlVersion,
+  }) async {
+    _logger.fine('getAllFaceMlResultFileIDsProcessedWithThumbnailOnly called');
+    final db = await instance.database;
+
+    String? whereString;
+    List<dynamic>? whereArgs;
+
+    if (mlVersion != null) {
+      whereString = '$mlVersionColumn = ?';
+      whereArgs = [mlVersion];
+    }
+
+    final List<Map<String, Object?>> results = await db.query(
+      facesTable,
+      where: whereString,
+      whereArgs: whereArgs,
+      orderBy: fileIDColumn,
+    );
+
+    return results
+        .map(
+          (result) =>
+              FaceMlResult.fromJsonString(result[faceMlResultColumn] as String),
+        )
+        .where((element) => element.onlyThumbnailUsed)
+        .map((result) => result.fileId)
+        .toSet();
+  }
+
+  /// Updates the faceMlResult for the given [faceMlResult.fileId]. Update is done regardless of the [faceMlResult.mlVersion].
+  /// However, if [updateHigherVersionOnly] is set to true, the update is only done if the [faceMlResult.mlVersion] is higher than the existing one.
+  Future<int> updateFaceMlResult(
+    FaceMlResult faceMlResult, {
+    bool updateHigherVersionOnly = false,
+  }) async {
+    _logger.fine('updateFaceMlResult called');
+
+    if (updateHigherVersionOnly) {
+      final existingResult = await getFaceMlResult(faceMlResult.fileId);
+      if (existingResult != null) {
+        if (faceMlResult.mlVersion <= existingResult.mlVersion) {
+          _logger.fine(
+            'FaceMlResult with file ID ${faceMlResult.fileId} already exists with equal or higher version. Skipping update.',
+          );
+          return 0;
+        }
+      }
+    }
+
+    final db = await instance.database;
+    return await db.update(
+      facesTable,
+      {
+        fileIDColumn: faceMlResult.fileId,
+        faceMlResultColumn: faceMlResult.toJsonString(),
+        mlVersionColumn: faceMlResult.mlVersion,
+      },
+      where: '$fileIDColumn = ?',
+      whereArgs: [faceMlResult.fileId],
+    );
+  }
+
+  Future<int> deleteFaceMlResult(int fileId) async {
+    _logger.fine('deleteFaceMlResult called');
+    final db = await instance.database;
+    final deleteCount = await db.delete(
+      facesTable,
+      where: '$fileIDColumn = ?',
+      whereArgs: [fileId],
+    );
+    _logger.fine('Deleted $deleteCount faceMlResults');
+    return deleteCount;
+  }
+
+  Future<void> createAllClusterResults(
+    List<ClusterResult> clusterResults, {
+    bool cleanExistingClusters = true,
+  }) async {
+    _logger.fine('createClusterResults called');
+    final db = await instance.database;
+
+    if (clusterResults.isEmpty) {
+      _logger.fine('No clusterResults given, skipping insert.');
+      return;
+    }
+
+    // Completely clean the table and start fresh
+    if (cleanExistingClusters) {
+      await deleteAllClusterResults();
+    }
+
+    // Insert all the cluster results
+    for (final clusterResult in clusterResults) {
+      await db.insert(
+        peopleTable,
+        {
+          personIDColumn: clusterResult.personId,
+          clusterResultColumn: clusterResult.toJsonString(),
+          centroidColumn: clusterResult.medoid.toString(),
+          centroidDistanceThresholdColumn:
+              clusterResult.medoidDistanceThreshold,
+        },
+        conflictAlgorithm: ConflictAlgorithm.replace,
+      );
+    }
+  }
+
+  Future<ClusterResult?> getClusterResult(int personId) async {
+    _logger.fine('getClusterResult called');
+    final db = await instance.database;
+
+    final result = await db.query(
+      peopleTable,
+      where: '$personIDColumn = ?',
+      whereArgs: [personId],
+      limit: 1,
+    );
+    if (result.isNotEmpty) {
+      return ClusterResult.fromJsonString(
+        result.first[clusterResultColumn] as String,
+      );
+    }
+    _logger.fine('No clusterResult found for personID $personId');
+    return null;
+  }
+
+  /// Returns the ClusterResult objects for the given [personIDs].
+  Future<List<ClusterResult>> getSelectedClusterResults(
+    List<int> personIDs,
+  ) async {
+    _logger.fine('getSelectedClusterResults called');
+    final db = await instance.database;
+
+    if (personIDs.isEmpty) {
+      _logger.warning('getSelectedClusterResults called with empty personIDs');
+      return <ClusterResult>[];
+    }
+
+    final results = await db.query(
+      peopleTable,
+      where: '$personIDColumn IN (${personIDs.join(',')})',
+      orderBy: personIDColumn,
+    );
+
+    return results
+        .map(
+          (result) => ClusterResult.fromJsonString(
+            result[clusterResultColumn] as String,
+          ),
+        )
+        .toList();
+  }
+
+  Future<List<ClusterResult>> getAllClusterResults() async {
+    _logger.fine('getAllClusterResults called');
+    final db = await instance.database;
+
+    final results = await db.query(
+      peopleTable,
+    );
+
+    return results
+        .map(
+          (result) => ClusterResult.fromJsonString(
+            result[clusterResultColumn] as String,
+          ),
+        )
+        .toList();
+  }
+
+  /// Returns the personIDs of all clustered people in the database.
+  Future<List<int>> getAllClusterIds() async {
+    _logger.fine('getAllClusterIds called');
+    final db = await instance.database;
+
+    final results = await db.query(
+      peopleTable,
+      columns: [personIDColumn],
+    );
+
+    return results.map((result) => result[personIDColumn] as int).toList();
+  }
+
+  /// Returns the fileIDs of all files associated with a given [personId].
+  Future<List<int>> getClusterFileIds(int personId) async {
+    _logger.fine('getClusterFileIds called');
+
+    final ClusterResult? clusterResult = await getClusterResult(personId);
+    if (clusterResult == null) {
+      return <int>[];
+    }
+    return clusterResult.uniqueFileIds;
+  }
+
+  Future<List<String>> getClusterFaceIds(int personId) async {
+    _logger.fine('getClusterFaceIds called');
+
+    final ClusterResult? clusterResult = await getClusterResult(personId);
+    if (clusterResult == null) {
+      return <String>[];
+    }
+    return clusterResult.faceIDs;
+  }
+
+  Future<List<Embedding>> getClusterEmbeddings(
+    int personId,
+  ) async {
+    _logger.fine('getClusterEmbeddings called');
+
+    final ClusterResult? clusterResult = await getClusterResult(personId);
+    if (clusterResult == null) return <Embedding>[];
+
+    final fileIds = clusterResult.uniqueFileIds;
+    final faceIds = clusterResult.faceIDs;
+    if (fileIds.length != faceIds.length) {
+      _logger.severe(
+        'fileIds and faceIds have different lengths: ${fileIds.length} vs ${faceIds.length}. This should not happen!',
+      );
+      return <Embedding>[];
+    }
+
+    final faceMlResults = await getSelectedFaceMlResults(fileIds);
+    if (faceMlResults.isEmpty) return <Embedding>[];
+
+    final embeddings = <Embedding>[];
+    for (var i = 0; i < faceMlResults.length; i++) {
+      final faceMlResult = faceMlResults[i];
+      final int faceIndex = faceMlResult.allFaceIds.indexOf(faceIds[i]);
+      if (faceIndex == -1) {
+        _logger.severe(
+          'Could not find faceIndex for faceId ${faceIds[i]} in faceMlResult ${faceMlResult.fileId}',
+        );
+        return <Embedding>[];
+      }
+      embeddings.add(faceMlResult.faces[faceIndex].embedding);
+    }
+
+    return embeddings;
+  }
+
+  Future<void> updateClusterResult(ClusterResult clusterResult) async {
+    _logger.fine('updateClusterResult called');
+    final db = await instance.database;
+    await db.update(
+      peopleTable,
+      {
+        personIDColumn: clusterResult.personId,
+        clusterResultColumn: clusterResult.toJsonString(),
+        centroidColumn: clusterResult.medoid.toString(),
+        centroidDistanceThresholdColumn: clusterResult.medoidDistanceThreshold,
+      },
+      where: '$personIDColumn = ?',
+      whereArgs: [clusterResult.personId],
+    );
+  }
+
+  Future<int> deleteClusterResult(int personId) async {
+    _logger.fine('deleteClusterResult called');
+    final db = await instance.database;
+    final deleteCount = await db.delete(
+      peopleTable,
+      where: '$personIDColumn = ?',
+      whereArgs: [personId],
+    );
+    _logger.fine('Deleted $deleteCount clusterResults');
+    return deleteCount;
+  }
+
+  Future<void> deleteAllClusterResults() async {
+    _logger.fine('deleteAllClusterResults called');
+    final db = await instance.database;
+    await db.execute(_deletePeopleTable);
+    await db.execute(createPeopleTable);
+  }
+
+  // TODO: current function implementation will skip inserting for a similar feedback, which means I can't remove two photos from the same person in a row
+  Future<void> createClusterFeedback<T extends ClusterFeedback>(
+    T feedback, {
+    bool skipIfSimilarFeedbackExists = false,
+  }) async {
+    _logger.fine('createClusterFeedback called');
+
+    // TODO: this skipping might cause issues for adding photos to the same person in a row!!
+    if (skipIfSimilarFeedbackExists &&
+        await doesSimilarClusterFeedbackExist(feedback)) {
+      _logger.fine(
+        'ClusterFeedback with ID ${feedback.feedbackID} already has a similar feedback installed. Skipping insert.',
+      );
+      return;
+    }
+
+    final db = await instance.database;
+    await db.insert(
+      feedbackTable,
+      {
+        feedbackIDColumn: feedback.feedbackID,
+        feedbackTypeColumn: feedback.typeString,
+        feedbackDataColumn: feedback.toJsonString(),
+        feedbackTimestampColumn: feedback.timestampString,
+        feedbackFaceMlVersionColumn: feedback.madeOnFaceMlVersion,
+        feedbackClusterMlVersionColumn: feedback.madeOnClusterMlVersion,
+      },
+      conflictAlgorithm: ConflictAlgorithm.replace,
+    );
+    return;
+  }
+
+  Future<bool> doesSimilarClusterFeedbackExist<T extends ClusterFeedback>(
+    T feedback,
+  ) async {
+    _logger.fine('doesClusterFeedbackExist called');
+
+    final List<T> existingFeedback =
+        await getAllClusterFeedback<T>(type: feedback.type);
+
+    if (existingFeedback.isNotEmpty) {
+      for (final existingFeedbackItem in existingFeedback) {
+        assert(
+          existingFeedbackItem.type == feedback.type,
+          'Feedback types should be the same!',
+        );
+        if (feedback.looselyMatchesMedoid(existingFeedbackItem)) {
+          _logger.fine(
+            'ClusterFeedback of type ${feedback.typeString} with ID ${feedback.feedbackID} already has a similar feedback installed!',
+          );
+          return true;
+        }
+      }
+    }
+    return false;
+  }
+
+  /// Returns all the clusterFeedbacks of type [T] which match the given [feedback], sorted by timestamp (latest first).
+  Future<List<T>> getAllMatchingClusterFeedback<T extends ClusterFeedback>(
+    T feedback, {
+    bool sortNewestFirst = true,
+  }) async {
+    _logger.fine('getAllMatchingClusterFeedback called');
+
+    final List<T> existingFeedback =
+        await getAllClusterFeedback<T>(type: feedback.type);
+    final List<T> matchingFeedback = <T>[];
+    if (existingFeedback.isNotEmpty) {
+      for (final existingFeedbackItem in existingFeedback) {
+        assert(
+          existingFeedbackItem.type == feedback.type,
+          'Feedback types should be the same!',
+        );
+        if (feedback.looselyMatchesMedoid(existingFeedbackItem)) {
+          _logger.fine(
+            'ClusterFeedback of type ${feedback.typeString} with ID ${feedback.feedbackID} already has a similar feedback installed!',
+          );
+          matchingFeedback.add(existingFeedbackItem);
+        }
+      }
+    }
+    if (sortNewestFirst) {
+      matchingFeedback.sort((a, b) => b.timestamp.compareTo(a.timestamp));
+    }
+    return matchingFeedback;
+  }
+
+  Future<List<T>> getAllClusterFeedback<T extends ClusterFeedback>({
+    required FeedbackType type,
+    int? mlVersion,
+    int? clusterMlVersion,
+  }) async {
+    _logger.fine('getAllClusterFeedback called');
+    final db = await instance.database;
+
+    // TODO: implement the versions for FeedbackType.imageFeedback and FeedbackType.faceFeedback and rename this function to getAllFeedback?
+
+    String whereString = '$feedbackTypeColumn = ?';
+    final List<dynamic> whereArgs = [type.toValueString()];
+
+    if (mlVersion != null) {
+      whereString += ' AND $feedbackFaceMlVersionColumn = ?';
+      whereArgs.add(mlVersion);
+    }
+    if (clusterMlVersion != null) {
+      whereString += ' AND $feedbackClusterMlVersionColumn = ?';
+      whereArgs.add(clusterMlVersion);
+    }
+
+    final results = await db.query(
+      feedbackTable,
+      where: whereString,
+      whereArgs: whereArgs,
+    );
+
+    if (results.isNotEmpty) {
+      if (ClusterFeedback.fromJsonStringRegistry.containsKey(type)) {
+        final Function(String) fromJsonString =
+            ClusterFeedback.fromJsonStringRegistry[type]!;
+        return results
+            .map((e) => fromJsonString(e[feedbackDataColumn] as String) as T)
+            .toList();
+      } else {
+        _logger.severe(
+          'No fromJsonString function found for type ${type.name}. This should not happen!',
+        );
+      }
+    }
+    _logger.fine(
+      'No clusterFeedback results found of type $type' +
+          (mlVersion != null ? ' and mlVersion $mlVersion' : '') +
+          (clusterMlVersion != null
+              ? ' and clusterMlVersion $clusterMlVersion'
+              : ''),
+    );
+    return <T>[];
+  }
+
+  Future<int> deleteClusterFeedback<T extends ClusterFeedback>(
+    T feedback,
+  ) async {
+    _logger.fine('deleteClusterFeedback called');
+    final db = await instance.database;
+    final deleteCount = await db.delete(
+      feedbackTable,
+      where: '$feedbackIDColumn = ?',
+      whereArgs: [feedback.feedbackID],
+    );
+    _logger.fine('Deleted $deleteCount clusterFeedbacks');
+    return deleteCount;
+  }
+}

mobile/lib/events/files_updated_event.dart

@@ -26,4 +26,5 @@ enum EventType {
   hide,
   unhide,
   coverChanged,
+  peopleChanged,
 }

mobile/lib/events/people_changed_event.dart

@@ -0,0 +1,3 @@
+import "package:photos/events/event.dart";
+
+class PeopleChangedEvent extends Event {}

mobile/lib/extensions/ml_linalg_extensions.dart

@@ -0,0 +1,193 @@
+import 'dart:math' as math show sin, cos, atan2, sqrt, pow;
+import 'package:ml_linalg/linalg.dart';
+
+extension SetVectorValues on Vector {
+  Vector setValues(int start, int end, Iterable<double> values) {
+    if (values.length > length) {
+      throw Exception('Values cannot be larger than vector');
+    } else if (end - start != values.length) {
+      throw Exception('Values must be same length as range');
+    } else if (start < 0 || end > length) {
+      throw Exception('Range must be within vector');
+    }
+    final tempList = toList();
+    tempList.replaceRange(start, end, values);
+    final newVector = Vector.fromList(tempList);
+    return newVector;
+  }
+}
+
+extension SetMatrixValues on Matrix {
+  Matrix setSubMatrix(
+    int startRow,
+    int endRow,
+    int startColumn,
+    int endColumn,
+    Iterable<Iterable<double>> values,
+  ) {
+    if (values.length > rowCount) {
+      throw Exception('New values cannot have more rows than original matrix');
+    } else if (values.elementAt(0).length > columnCount) {
+      throw Exception(
+        'New values cannot have more columns than original matrix',
+      );
+    } else if (endRow - startRow != values.length) {
+      throw Exception('Values (number of rows) must be same length as range');
+    } else if (endColumn - startColumn != values.elementAt(0).length) {
+      throw Exception(
+        'Values (number of columns) must be same length as range',
+      );
+    } else if (startRow < 0 ||
+        endRow > rowCount ||
+        startColumn < 0 ||
+        endColumn > columnCount) {
+      throw Exception('Range must be within matrix');
+    }
+    final tempList = asFlattenedList
+        .toList(); // You need `.toList()` here to make sure the list is growable, otherwise `replaceRange` will throw an error
+    for (var i = startRow; i < endRow; i++) {
+      tempList.replaceRange(
+        i * columnCount + startColumn,
+        i * columnCount + endColumn,
+        values.elementAt(i).toList(),
+      );
+    }
+    final newMatrix = Matrix.fromFlattenedList(tempList, rowCount, columnCount);
+    return newMatrix;
+  }
+
+  Matrix setValues(
+    int startRow,
+    int endRow,
+    int startColumn,
+    int endColumn,
+    Iterable<double> values,
+  ) {
+    if ((startRow - endRow) * (startColumn - endColumn) != values.length) {
+      throw Exception('Values must be same length as range');
+    } else if (startRow < 0 ||
+        endRow > rowCount ||
+        startColumn < 0 ||
+        endColumn > columnCount) {
+      throw Exception('Range must be within matrix');
+    }
+
+    final tempList = asFlattenedList
+        .toList(); // You need `.toList()` here to make sure the list is growable, otherwise `replaceRange` will throw an error
+    var index = 0;
+    for (var i = startRow; i < endRow; i++) {
+      for (var j = startColumn; j < endColumn; j++) {
+        tempList[i * columnCount + j] = values.elementAt(index);
+        index++;
+      }
+    }
+    final newMatrix = Matrix.fromFlattenedList(tempList, rowCount, columnCount);
+    return newMatrix;
+  }
+
+  Matrix setValue(int row, int column, double value) {
+    if (row < 0 || row > rowCount || column < 0 || column > columnCount) {
+      throw Exception('Index must be within range of matrix');
+    }
+    final tempList = asFlattenedList;
+    tempList[row * columnCount + column] = value;
+    final newMatrix = Matrix.fromFlattenedList(tempList, rowCount, columnCount);
+    return newMatrix;
+  }
+
+  Matrix appendRow(List<double> row) {
+    final oldNumberOfRows = rowCount;
+    final oldNumberOfColumns = columnCount;
+    if (row.length != oldNumberOfColumns) {
+      throw Exception('Row must have same number of columns as matrix');
+    }
+    final flatListMatrix = asFlattenedList;
+    flatListMatrix.addAll(row);
+    return Matrix.fromFlattenedList(
+      flatListMatrix,
+      oldNumberOfRows + 1,
+      oldNumberOfColumns,
+    );
+  }
+}
+
+extension MatrixCalculations on Matrix {
+  double determinant() {
+    final int length = rowCount;
+    if (length != columnCount) {
+      throw Exception('Matrix must be square');
+    }
+    if (length == 1) {
+      return this[0][0];
+    } else if (length == 2) {
+      return this[0][0] * this[1][1] - this[0][1] * this[1][0];
+    } else {
+      throw Exception('Determinant for Matrix larger than 2x2 not implemented');
+    }
+  }
+
+  /// Computes the singular value decomposition of a matrix, using https://lucidar.me/en/mathematics/singular-value-decomposition-of-a-2x2-matrix/ as reference, but with slightly different signs for the second columns of U and V
+  Map<String, dynamic> svd() {
+    if (rowCount != 2 || columnCount != 2) {
+      throw Exception('Matrix must be 2x2');
+    }
+    final a = this[0][0];
+    final b = this[0][1];
+    final c = this[1][0];
+    final d = this[1][1];
+
+    // Computation of U matrix
+    final tempCalc = a * a + b * b - c * c - d * d;
+    final theta = 0.5 * math.atan2(2 * a * c + 2 * b * d, tempCalc);
+    final U = Matrix.fromList([
+      [math.cos(theta), math.sin(theta)],
+      [math.sin(theta), -math.cos(theta)],
+    ]);
+
+    // Computation of S matrix
+    // ignore: non_constant_identifier_names
+    final S1 = a * a + b * b + c * c + d * d;
+    // ignore: non_constant_identifier_names
+    final S2 =
+        math.sqrt(math.pow(tempCalc, 2) + 4 * math.pow(a * c + b * d, 2));
+    final sigma1 = math.sqrt((S1 + S2) / 2);
+    final sigma2 = math.sqrt((S1 - S2) / 2);
+    final S = Vector.fromList([sigma1, sigma2]);
+
+    // Computation of V matrix
+    final tempCalc2 = a * a - b * b + c * c - d * d;
+    final phi = 0.5 * math.atan2(2 * a * b + 2 * c * d, tempCalc2);
+    final s11 = (a * math.cos(theta) + c * math.sin(theta)) * math.cos(phi) +
+        (b * math.cos(theta) + d * math.sin(theta)) * math.sin(phi);
+    final s22 = (a * math.sin(theta) - c * math.cos(theta)) * math.sin(phi) +
+        (-b * math.sin(theta) + d * math.cos(theta)) * math.cos(phi);
+    final V = Matrix.fromList([
+      [s11.sign * math.cos(phi), s22.sign * math.sin(phi)],
+      [s11.sign * math.sin(phi), -s22.sign * math.cos(phi)],
+    ]);
+
+    return {
+      'U': U,
+      'S': S,
+      'V': V,
+    };
+  }
+
+  int matrixRank() {
+    final svdResult = svd();
+    final Vector S = svdResult['S']!;
+    final rank = S.toList().where((element) => element > 1e-10).length;
+    return rank;
+  }
+}
+
+extension TransformMatrix on Matrix {
+  List<List<double>> to2DList() {
+    final List<List<double>> outerList = [];
+    for (var i = 0; i < rowCount; i++) {
+      final innerList = this[i].toList();
+      outerList.add(innerList);
+    }
+    return outerList;
+  }
+}

mobile/lib/face/db.dart

@@ -0,0 +1,679 @@
+import 'dart:async';
+import "dart:math";
+import "dart:typed_data";
+
+import "package:collection/collection.dart";
+import "package:flutter/foundation.dart";
+import 'package:logging/logging.dart';
+import 'package:path/path.dart' show join;
+import 'package:path_provider/path_provider.dart';
+import 'package:photos/face/db_fields.dart';
+import "package:photos/face/db_model_mappers.dart";
+import "package:photos/face/model/face.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/models/file/file.dart";
+import "package:photos/services/face_ml/blur_detection/blur_constants.dart";
+import 'package:sqflite/sqflite.dart';
+
+/// Stores all data for the ML-related features. The database can be accessed by `MlDataDB.instance.database`.
+///
+/// This includes:
+/// [facesTable] - Stores all the detected faces and its embeddings in the images.
+/// [peopleTable] - Stores all the clusters of faces which are considered to be the same person.
+class FaceMLDataDB {
+  static final Logger _logger = Logger("FaceMLDataDB");
+
+  static const _databaseName = "ente.face_ml_db.db";
+  static const _databaseVersion = 1;
+
+  FaceMLDataDB._privateConstructor();
+
+  static final FaceMLDataDB instance = FaceMLDataDB._privateConstructor();
+
+  static Future<Database>? _dbFuture;
+
+  Future<Database> get database async {
+    _dbFuture ??= _initDatabase();
+    return _dbFuture!;
+  }
+
+  Future<Database> _initDatabase() async {
+    final documentsDirectory = await getApplicationDocumentsDirectory();
+    final String databaseDirectory =
+        join(documentsDirectory.path, _databaseName);
+    return await openDatabase(
+      databaseDirectory,
+      version: _databaseVersion,
+      onCreate: _onCreate,
+    );
+  }
+
+  Future _onCreate(Database db, int version) async {
+    await db.execute(createFacesTable);
+    await db.execute(createPeopleTable);
+    await db.execute(createClusterTable);
+    await db.execute(createClusterSummaryTable);
+    await db.execute(createNotPersonFeedbackTable);
+  }
+
+  // bulkInsertFaces inserts the faces in the database in batches of 1000.
+  // This is done to avoid the error "too many SQL variables" when inserting
+  // a large number of faces.
+  Future<void> bulkInsertFaces(List<Face> faces) async {
+    final db = await instance.database;
+    const batchSize = 500;
+    final numBatches = (faces.length / batchSize).ceil();
+    for (int i = 0; i < numBatches; i++) {
+      final start = i * batchSize;
+      final end = min((i + 1) * batchSize, faces.length);
+      final batch = faces.sublist(start, end);
+      final batchInsert = db.batch();
+      for (final face in batch) {
+        batchInsert.insert(
+          facesTable,
+          mapRemoteToFaceDB(face),
+          conflictAlgorithm: ConflictAlgorithm.ignore,
+        );
+      }
+      await batchInsert.commit(noResult: true);
+    }
+  }
+
+  Future<Set<int>> getIndexedFileIds() async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT DISTINCT $fileIDColumn FROM $facesTable',
+    );
+    return maps.map((e) => e[fileIDColumn] as int).toSet();
+  }
+
+  Future<Map<int, int>> clusterIdToFaceCount() async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $cluserIDColumn, COUNT(*) as count FROM $facesTable where $cluserIDColumn IS NOT NULL GROUP BY $cluserIDColumn ',
+    );
+    final Map<int, int> result = {};
+    for (final map in maps) {
+      result[map[cluserIDColumn] as int] = map['count'] as int;
+    }
+    return result;
+  }
+
+  Future<Set<int>> getPersonIgnoredClusters(String personID) async {
+    final db = await instance.database;
+    // find out clusterIds that are assigned to other persons using the clusters table
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $cluserIDColumn FROM $clustersTable WHERE $personIdColumn != ? AND $personIdColumn IS NOT NULL',
+      [personID],
+    );
+    final Set<int> ignoredClusterIDs =
+        maps.map((e) => e[cluserIDColumn] as int).toSet();
+    final List<Map<String, dynamic>> rejectMaps = await db.rawQuery(
+      'SELECT $cluserIDColumn FROM $notPersonFeedback WHERE $personIdColumn = ?',
+      [personID],
+    );
+    final Set<int> rejectClusterIDs =
+        rejectMaps.map((e) => e[cluserIDColumn] as int).toSet();
+    return ignoredClusterIDs.union(rejectClusterIDs);
+  }
+
+  Future<Set<int>> getPersonClusterIDs(String personID) async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $cluserIDColumn FROM $clustersTable WHERE $personIdColumn = ?',
+      [personID],
+    );
+    return maps.map((e) => e[cluserIDColumn] as int).toSet();
+  }
+
+  Future<void> clearTable() async {
+    final db = await instance.database;
+    await db.delete(facesTable);
+    await db.delete(createClusterTable);
+    await db.delete(clusterSummaryTable);
+    await db.delete(peopleTable);
+    await db.delete(notPersonFeedback);
+  }
+
+  Future<Iterable<Uint8List>> getFaceEmbeddingsForCluster(
+    int clusterID, {
+    int? limit,
+  }) async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $faceEmbeddingBlob FROM $facesTable WHERE $cluserIDColumn = ? ${limit != null ? 'LIMIT $limit' : ''}',
+      [clusterID],
+    );
+    return maps.map((e) => e[faceEmbeddingBlob] as Uint8List);
+  }
+
+  Future<Map<int, int>> getFileIdToCount() async {
+    final Map<int, int> result = {};
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $fileIDColumn, COUNT(*) as count FROM $facesTable where $faceScore > 0.8 GROUP BY $fileIDColumn',
+    );
+
+    for (final map in maps) {
+      result[map[fileIDColumn] as int] = map['count'] as int;
+    }
+    return result;
+  }
+
+  Future<Face?> getCoverFaceForPerson({
+    required int recentFileID,
+    String? personID,
+    int? clusterID,
+  }) async {
+    // read person from db
+    final db = await instance.database;
+    if (personID != null) {
+      final List<Map<String, dynamic>> maps = await db.rawQuery(
+        'SELECT * FROM $peopleTable where $idColumn = ?',
+        [personID],
+      );
+      if (maps.isEmpty) {
+        throw Exception("Person with id $personID not found");
+      }
+
+      final person = mapRowToPerson(maps.first);
+      final List<int> fileId = [recentFileID];
+      int? avatarFileId;
+      if (person.attr.avatarFaceId != null) {
+        avatarFileId = int.tryParse(person.attr.avatarFaceId!.split('-')[0]);
+        if (avatarFileId != null) {
+          fileId.add(avatarFileId);
+        }
+      }
+      final cluterRows = await db.query(
+        clustersTable,
+        columns: [cluserIDColumn],
+        where: '$personIdColumn = ?',
+        whereArgs: [personID],
+      );
+      final clusterIDs =
+          cluterRows.map((e) => e[cluserIDColumn] as int).toList();
+      final List<Map<String, dynamic>> faceMaps = await db.rawQuery(
+        'SELECT * FROM $facesTable where $faceClusterId IN (${clusterIDs.join(",")}) AND $fileIDColumn in (${fileId.join(",")}) AND $faceScore > 0.8 ORDER BY $faceScore DESC',
+      );
+      if (faceMaps.isNotEmpty) {
+        if (avatarFileId != null) {
+          final row = faceMaps.firstWhereOrNull(
+            (element) => (element[fileIDColumn] as int) == avatarFileId,
+          );
+          if (row != null) {
+            return mapRowToFace(row);
+          }
+        }
+        return mapRowToFace(faceMaps.first);
+      }
+    }
+    if (clusterID != null) {
+      final clusterIDs = [clusterID];
+      final List<Map<String, dynamic>> faceMaps = await db.rawQuery(
+        'SELECT * FROM $facesTable where $faceClusterId IN (${clusterIDs.join(",")}) AND $fileIDColumn = $recentFileID ',
+      );
+      if (faceMaps.isNotEmpty) {
+        return mapRowToFace(faceMaps.first);
+      }
+    }
+    if (personID == null && clusterID == null) {
+      throw Exception("personID and clusterID cannot be null");
+    }
+    return null;
+  }
+
+  Future<List<Face>> getFacesForGivenFileID(int fileUploadID) async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.query(
+      facesTable,
+      columns: [
+        fileIDColumn,
+        faceIDColumn,
+        faceDetectionColumn,
+        faceEmbeddingBlob,
+        faceScore,
+        faceBlur,
+        faceClusterId,
+        faceClosestDistColumn,
+        faceClosestFaceID,
+        faceConfirmedColumn,
+        mlVersionColumn,
+      ],
+      where: '$fileIDColumn = ?',
+      whereArgs: [fileUploadID],
+    );
+    return maps.map((e) => mapRowToFace(e)).toList();
+  }
+
+  Future<Face?> getFaceForFaceID(String faceID) async {
+    final db = await instance.database;
+    final result = await db.rawQuery(
+      'SELECT * FROM $facesTable where $faceIDColumn = ?',
+      [faceID],
+    );
+    if (result.isEmpty) {
+      return null;
+    }
+    return mapRowToFace(result.first);
+  }
+
+  Future<Map<String, int?>> getFaceIdsToClusterIds(
+    Iterable<String> faceIds,
+  ) async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $faceIDColumn, $faceClusterId FROM $facesTable where $faceIDColumn IN (${faceIds.map((id) => "'$id'").join(",")})',
+    );
+    final Map<String, int?> result = {};
+    for (final map in maps) {
+      result[map[faceIDColumn] as String] = map[faceClusterId] as int?;
+    }
+    return result;
+  }
+
+  Future<Map<int, Set<int>>> getFileIdToClusterIds() async {
+    final Map<int, Set<int>> result = {};
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $faceClusterId, $fileIDColumn FROM $facesTable where $faceClusterId IS NOT NULL',
+    );
+
+    for (final map in maps) {
+      final personID = map[faceClusterId] as int;
+      final fileID = map[fileIDColumn] as int;
+      result[fileID] = (result[fileID] ?? {})..add(personID);
+    }
+    return result;
+  }
+
+  Future<void> updatePersonIDForFaceIDIFNotSet(
+    Map<String, int> faceIDToPersonID,
+  ) async {
+    final db = await instance.database;
+
+    // Start a batch
+    final batch = db.batch();
+
+    for (final map in faceIDToPersonID.entries) {
+      final faceID = map.key;
+      final personID = map.value;
+      batch.update(
+        facesTable,
+        {faceClusterId: personID},
+        where: '$faceIDColumn = ? AND $faceClusterId IS NULL',
+        whereArgs: [faceID],
+      );
+    }
+    // Commit the batch
+    await batch.commit(noResult: true);
+  }
+
+  Future<void> forceUpdateClusterIds(
+    Map<String, int> faceIDToPersonID,
+  ) async {
+    final db = await instance.database;
+
+    // Start a batch
+    final batch = db.batch();
+
+    for (final map in faceIDToPersonID.entries) {
+      final faceID = map.key;
+      final personID = map.value;
+      batch.update(
+        facesTable,
+        {faceClusterId: personID},
+        where: '$faceIDColumn = ?',
+        whereArgs: [faceID],
+      );
+    }
+    // Commit the batch
+    await batch.commit(noResult: true);
+  }
+
+  /// Returns a map of faceID to record of faceClusterID and faceEmbeddingBlob
+  ///
+  /// Only selects faces with score greater than [minScore] and blur score greater than [minClarity]
+  Future<Map<String, (int?, Uint8List)>> getFaceEmbeddingMap({
+    double minScore = 0.8,
+    int minClarity = kLaplacianThreshold,
+    int maxRows = 20000,
+  }) async {
+    _logger.info('reading as float');
+    final db = await instance.database;
+
+    // Define the batch size
+    const batchSize = 10000;
+    int offset = 0;
+
+    final Map<String, (int?, Uint8List)> result = {};
+    while (true) {
+      // Query a batch of rows
+      final List<Map<String, dynamic>> maps = await db.query(
+        facesTable,
+        columns: [faceIDColumn, faceClusterId, faceEmbeddingBlob],
+        where: '$faceScore > $minScore and $faceBlur > $minClarity',
+        limit: batchSize,
+        offset: offset,
+        // orderBy: '$faceClusterId DESC',
+        orderBy: '$faceIDColumn DESC',
+      );
+      // Break the loop if no more rows
+      if (maps.isEmpty) {
+        break;
+      }
+      for (final map in maps) {
+        final faceID = map[faceIDColumn] as String;
+        result[faceID] =
+            (map[faceClusterId] as int?, map[faceEmbeddingBlob] as Uint8List);
+      }
+      if (result.length >= 20000) {
+        break;
+      }
+      offset += batchSize;
+    }
+    return result;
+  }
+
+  Future<Map<String, Uint8List>> getFaceEmbeddingMapForFile(
+    List<int> fileIDs,
+  ) async {
+    _logger.info('reading as float');
+    final db = await instance.database;
+
+    // Define the batch size
+    const batchSize = 10000;
+    int offset = 0;
+
+    final Map<String, Uint8List> result = {};
+    while (true) {
+      // Query a batch of rows
+      final List<Map<String, dynamic>> maps = await db.query(
+        facesTable,
+        columns: [faceIDColumn, faceEmbeddingBlob],
+        where:
+            '$faceScore > 0.8 AND $faceBlur > $kLaplacianThreshold AND $fileIDColumn IN (${fileIDs.join(",")})',
+        limit: batchSize,
+        offset: offset,
+        orderBy: '$faceIDColumn DESC',
+      );
+      // Break the loop if no more rows
+      if (maps.isEmpty) {
+        break;
+      }
+      for (final map in maps) {
+        final faceID = map[faceIDColumn] as String;
+        result[faceID] = map[faceEmbeddingBlob] as Uint8List;
+      }
+      if (result.length > 10000) {
+        break;
+      }
+      offset += batchSize;
+    }
+    return result;
+  }
+
+  Future<void> resetClusterIDs() async {
+    final db = await instance.database;
+    await db.update(
+      facesTable,
+      {faceClusterId: null},
+    );
+  }
+
+  Future<void> insert(Person p, int cluserID) async {
+    debugPrint("inserting person");
+    final db = await instance.database;
+    await db.insert(
+      peopleTable,
+      mapPersonToRow(p),
+      conflictAlgorithm: ConflictAlgorithm.replace,
+    );
+    await db.insert(
+      clustersTable,
+      {
+        personIdColumn: p.remoteID,
+        cluserIDColumn: cluserID,
+      },
+      conflictAlgorithm: ConflictAlgorithm.replace,
+    );
+  }
+
+  Future<void> updatePerson(Person p) async {
+    final db = await instance.database;
+    await db.update(
+      peopleTable,
+      mapPersonToRow(p),
+      where: '$idColumn = ?',
+      whereArgs: [p.remoteID],
+    );
+  }
+
+  Future<void> assignClusterToPerson({
+    required String personID,
+    required int clusterID,
+  }) async {
+    final db = await instance.database;
+    await db.insert(
+      clustersTable,
+      {
+        personIdColumn: personID,
+        cluserIDColumn: clusterID,
+      },
+    );
+  }
+
+  Future<void> captureNotPersonFeedback({
+    required String personID,
+    required int clusterID,
+  }) async {
+    final db = await instance.database;
+    await db.insert(
+      notPersonFeedback,
+      {
+        personIdColumn: personID,
+        cluserIDColumn: clusterID,
+      },
+    );
+  }
+
+  Future<int> removeClusterToPerson({
+    required String personID,
+    required int clusterID,
+  }) async {
+    final db = await instance.database;
+    return db.delete(
+      clustersTable,
+      where: '$personIdColumn = ? AND $cluserIDColumn = ?',
+      whereArgs: [personID, clusterID],
+    );
+  }
+
+  // for a given personID, return a map of clusterID to fileIDs using join query
+  Future<Map<int, Set<int>>> getFileIdToClusterIDSet(String personID) {
+    final db = instance.database;
+    return db.then((db) async {
+      final List<Map<String, dynamic>> maps = await db.rawQuery(
+        'SELECT $clustersTable.$cluserIDColumn, $fileIDColumn FROM $facesTable '
+        'INNER JOIN $clustersTable '
+        'ON $facesTable.$faceClusterId = $clustersTable.$cluserIDColumn '
+        'WHERE $clustersTable.$personIdColumn = ?',
+        [personID],
+      );
+      final Map<int, Set<int>> result = {};
+      for (final map in maps) {
+        final clusterID = map[cluserIDColumn] as int;
+        final fileID = map[fileIDColumn] as int;
+        result[fileID] = (result[fileID] ?? {})..add(clusterID);
+      }
+      return result;
+    });
+  }
+
+  Future<Map<int, Set<int>>> getFileIdToClusterIDSetForCluster(
+    Set<int> clusterIDs,
+  ) {
+    final db = instance.database;
+    return db.then((db) async {
+      final List<Map<String, dynamic>> maps = await db.rawQuery(
+        'SELECT $cluserIDColumn, $fileIDColumn FROM $facesTable '
+        'WHERE $cluserIDColumn IN (${clusterIDs.join(",")})',
+      );
+      final Map<int, Set<int>> result = {};
+      for (final map in maps) {
+        final clusterID = map[cluserIDColumn] as int;
+        final fileID = map[fileIDColumn] as int;
+        result[fileID] = (result[fileID] ?? {})..add(clusterID);
+      }
+      return result;
+    });
+  }
+
+  Future<void> clusterSummaryUpdate(Map<int, (Uint8List, int)> summary) async {
+    final db = await instance.database;
+    var batch = db.batch();
+    int batchCounter = 0;
+    for (final entry in summary.entries) {
+      if (batchCounter == 400) {
+        await batch.commit(noResult: true);
+        batch = db.batch();
+        batchCounter = 0;
+      }
+      final int cluserID = entry.key;
+      final int count = entry.value.$2;
+      final Uint8List avg = entry.value.$1;
+      batch.insert(
+        clusterSummaryTable,
+        {
+          cluserIDColumn: cluserID,
+          avgColumn: avg,
+          countColumn: count,
+        },
+        conflictAlgorithm: ConflictAlgorithm.replace,
+      );
+      batchCounter++;
+    }
+    await batch.commit(noResult: true);
+  }
+
+  /// Returns a map of clusterID to (avg embedding, count)
+  Future<Map<int, (Uint8List, int)>> clusterSummaryAll() async {
+    final db = await instance.database;
+    final Map<int, (Uint8List, int)> result = {};
+    final rows = await db.rawQuery('SELECT * from $clusterSummaryTable');
+    for (final r in rows) {
+      final id = r[cluserIDColumn] as int;
+      final avg = r[avgColumn] as Uint8List;
+      final count = r[countColumn] as int;
+      result[id] = (avg, count);
+    }
+    return result;
+  }
+
+  Future<Map<int, String>> getCluserIDToPersonMap() async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $personIdColumn, $cluserIDColumn FROM $clustersTable',
+    );
+    final Map<int, String> result = {};
+    for (final map in maps) {
+      result[map[cluserIDColumn] as int] = map[personIdColumn] as String;
+    }
+    return result;
+  }
+
+  Future<(Map<int, Person>, Map<String, Person>)> getClusterIdToPerson() async {
+    final db = await instance.database;
+    final Map<String, Person> peopleMap = await getPeopleMap();
+    final List<Map<String, dynamic>> maps = await db.rawQuery(
+      'SELECT $personIdColumn, $cluserIDColumn FROM $clustersTable',
+    );
+
+    final Map<int, Person> result = {};
+    for (final map in maps) {
+      final Person? p = peopleMap[map[personIdColumn] as String];
+      if (p != null) {
+        result[map[cluserIDColumn] as int] = p;
+      } else {
+        _logger.warning(
+          'Person with id ${map[personIdColumn]} not found',
+        );
+      }
+    }
+    return (result, peopleMap);
+  }
+
+  Future<Map<String, Person>> getPeopleMap() async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.query(
+      peopleTable,
+      columns: [
+        idColumn,
+        nameColumn,
+        personHiddenColumn,
+        clusterToFaceIdJson,
+        coverFaceIDColumn,
+      ],
+    );
+    final Map<String, Person> result = {};
+    for (final map in maps) {
+      result[map[idColumn] as String] = mapRowToPerson(map);
+    }
+    return result;
+  }
+
+  Future<List<Person>> getPeople() async {
+    final db = await instance.database;
+    final List<Map<String, dynamic>> maps = await db.query(
+      peopleTable,
+      columns: [
+        idColumn,
+        nameColumn,
+        personHiddenColumn,
+        clusterToFaceIdJson,
+        coverFaceIDColumn,
+      ],
+    );
+    return maps.map((map) => mapRowToPerson(map)).toList();
+  }
+
+  /// WARNING: This will delete ALL data in the database! Only use this for debug/testing purposes!
+  Future<void> dropClustersAndPeople({bool faces = false}) async {
+    final db = await instance.database;
+    if (faces) {
+      await db.execute(deleteFacesTable);
+      await db.execute(createFacesTable);
+    }
+    await db.execute(deletePeopleTable);
+    await db.execute(dropClustersTable);
+    await db.execute(dropClusterSummaryTable);
+    await db.execute(dropNotPersonFeedbackTable);
+
+    // await db.execute(createFacesTable);
+    await db.execute(createPeopleTable);
+    await db.execute(createClusterTable);
+    await db.execute(createNotPersonFeedbackTable);
+    await db.execute(createClusterSummaryTable);
+  }
+
+  Future<void> removePersonFromFiles(List<EnteFile> files, Person p) async {
+    final db = await instance.database;
+    final result = await db.rawQuery(
+      'SELECT $faceIDColumn FROM $facesTable LEFT JOIN $clustersTable '
+      'ON $facesTable.$faceClusterId = $clustersTable.$cluserIDColumn '
+      'WHERE $clustersTable.$personIdColumn = ? AND $facesTable.$fileIDColumn IN (${files.map((e) => e.uploadedFileID).join(",")})',
+      [p.remoteID],
+    );
+    // get max clusterID
+    final maxRows =
+        await db.rawQuery('SELECT max($faceClusterId) from $facesTable');
+    int maxClusterID = maxRows.first.values.first as int;
+    final Map<String, int> faceIDToClusterID = {};
+    for (final faceRow in result) {
+      final faceID = faceRow[faceIDColumn] as String;
+      faceIDToClusterID[faceID] = maxClusterID + 1;
+      maxClusterID = maxClusterID + 1;
+    }
+    await forceUpdateClusterIds(faceIDToClusterID);
+  }
+}

mobile/lib/face/db_fields.dart

@@ -0,0 +1,99 @@
+// Faces Table Fields & Schema Queries
+import "package:photos/services/face_ml/blur_detection/blur_constants.dart";
+
+const facesTable = 'faces';
+const fileIDColumn = 'file_id';
+const faceIDColumn = 'face_id';
+const faceDetectionColumn = 'detection';
+const faceEmbeddingBlob = 'eBlob';
+const faceScore = 'score';
+const faceBlur = 'blur';
+const faceClusterId = 'cluster_id';
+const faceConfirmedColumn = 'confirmed';
+const faceClosestDistColumn = 'close_dist';
+const faceClosestFaceID = 'close_face_id';
+const mlVersionColumn = 'ml_version';
+
+const createFacesTable = '''CREATE TABLE IF NOT EXISTS $facesTable (
+  $fileIDColumn	INTEGER NOT NULL,
+  $faceIDColumn  TEXT NOT NULL,
+	$faceDetectionColumn	TEXT NOT NULL,
+  $faceEmbeddingBlob BLOB NOT NULL,
+  $faceScore  REAL NOT NULL,
+  $faceBlur REAL NOT NULL DEFAULT $kLapacianDefault,
+	$faceClusterId	INTEGER,
+	$faceClosestDistColumn	REAL,
+  $faceClosestFaceID  TEXT,
+	$faceConfirmedColumn  INTEGER NOT NULL DEFAULT 0,
+  $mlVersionColumn	INTEGER NOT NULL DEFAULT -1,
+  PRIMARY KEY($fileIDColumn, $faceIDColumn)
+  );
+  ''';
+
+const deleteFacesTable = 'DROP TABLE IF EXISTS $facesTable';
+// End of Faces Table Fields & Schema Queries
+
+// People Table Fields & Schema Queries
+const peopleTable = 'people';
+const idColumn = 'id';
+const nameColumn = 'name';
+const personHiddenColumn = 'hidden';
+const clusterToFaceIdJson = 'clusterToFaceIds';
+const coverFaceIDColumn = 'cover_face_id';
+
+const createPeopleTable = '''CREATE TABLE IF NOT EXISTS $peopleTable (
+  $idColumn	TEXT NOT NULL UNIQUE,
+	$nameColumn	TEXT NOT NULL DEFAULT '',
+  $personHiddenColumn	INTEGER NOT NULL DEFAULT 0,
+  $clusterToFaceIdJson	TEXT NOT NULL DEFAULT '{}',
+  $coverFaceIDColumn	TEXT,
+	PRIMARY KEY($idColumn)
+  );
+  ''';
+
+const deletePeopleTable = 'DROP TABLE IF EXISTS $peopleTable';
+//End People Table Fields & Schema Queries
+
+// Clusters Table Fields & Schema Queries
+const clustersTable = 'clusters';
+const personIdColumn = 'person_id';
+const cluserIDColumn = 'cluster_id';
+
+const createClusterTable = '''
+CREATE TABLE IF NOT EXISTS $clustersTable (
+  $personIdColumn	TEXT NOT NULL,
+  $cluserIDColumn	INTEGER NOT NULL,
+  PRIMARY KEY($personIdColumn, $cluserIDColumn)
+);
+''';
+const dropClustersTable = 'DROP TABLE IF EXISTS $clustersTable';
+// End Clusters Table Fields & Schema Queries
+
+/// Cluster Summary Table Fields & Schema Queries
+const clusterSummaryTable = 'cluster_summary';
+const avgColumn = 'avg';
+const countColumn = 'count';
+const createClusterSummaryTable = '''
+CREATE TABLE IF NOT EXISTS $clusterSummaryTable (
+  $cluserIDColumn	INTEGER NOT NULL,
+  $avgColumn BLOB NOT NULL,
+  $countColumn INTEGER NOT NULL,
+  PRIMARY KEY($cluserIDColumn)
+);
+''';
+
+const dropClusterSummaryTable = 'DROP TABLE IF EXISTS $clusterSummaryTable';
+
+/// End Cluster Summary Table Fields & Schema Queries
+
+/// notPersonFeedback Table Fields & Schema Queries
+const notPersonFeedback = 'not_person_feedback';
+
+const createNotPersonFeedbackTable = '''
+CREATE TABLE IF NOT EXISTS $notPersonFeedback (
+  $personIdColumn	TEXT NOT NULL,
+  $cluserIDColumn	INTEGER NOT NULL
+);
+''';
+const dropNotPersonFeedbackTable = 'DROP TABLE IF EXISTS $notPersonFeedback';
+// End Clusters Table Fields & Schema Queries

mobile/lib/face/db_model_mappers.dart

@@ -0,0 +1,86 @@
+import "dart:convert";
+
+import 'package:photos/face/db_fields.dart';
+import "package:photos/face/model/detection.dart";
+import "package:photos/face/model/face.dart";
+import "package:photos/face/model/person.dart";
+import 'package:photos/face/model/person_face.dart';
+import "package:photos/generated/protos/ente/common/vector.pb.dart";
+
+int boolToSQLInt(bool? value, {bool defaultValue = false}) {
+  final bool v = value ?? defaultValue;
+  if (v == false) {
+    return 0;
+  } else {
+    return 1;
+  }
+}
+
+bool sqlIntToBool(int? value, {bool defaultValue = false}) {
+  final int v = value ?? (defaultValue ? 1 : 0);
+  if (v == 0) {
+    return false;
+  } else {
+    return true;
+  }
+}
+
+Map<String, dynamic> mapToFaceDB(PersonFace personFace) {
+  return {
+    faceIDColumn: personFace.face.faceID,
+    faceDetectionColumn: json.encode(personFace.face.detection.toJson()),
+    faceConfirmedColumn: boolToSQLInt(personFace.confirmed),
+    faceClusterId: personFace.personID,
+    faceClosestDistColumn: personFace.closeDist,
+    faceClosestFaceID: personFace.closeFaceID,
+  };
+}
+
+Map<String, dynamic> mapPersonToRow(Person p) {
+  return {
+    idColumn: p.remoteID,
+    nameColumn: p.attr.name,
+    personHiddenColumn: boolToSQLInt(p.attr.isHidden),
+    coverFaceIDColumn: p.attr.avatarFaceId,
+    clusterToFaceIdJson: jsonEncode(p.attr.faces.toList()),
+  };
+}
+
+Person mapRowToPerson(Map<String, dynamic> row) {
+  return Person(
+    row[idColumn] as String,
+    PersonAttr(
+      name: row[nameColumn] as String,
+      isHidden: sqlIntToBool(row[personHiddenColumn] as int),
+      avatarFaceId: row[coverFaceIDColumn] as String?,
+      faces: (jsonDecode(row[clusterToFaceIdJson]) as List)
+          .map((e) => e.toString())
+          .toList(),
+    ),
+  );
+}
+
+Map<String, dynamic> mapRemoteToFaceDB(Face face) {
+  return {
+    faceIDColumn: face.faceID,
+    fileIDColumn: face.fileID,
+    faceDetectionColumn: json.encode(face.detection.toJson()),
+    faceEmbeddingBlob: EVector(
+      values: face.embedding,
+    ).writeToBuffer(),
+    faceScore: face.score,
+    faceBlur: face.blur,
+    mlVersionColumn: 1,
+  };
+}
+
+Face mapRowToFace(Map<String, dynamic> row) {
+  return Face(
+    row[faceIDColumn] as String,
+    row[fileIDColumn] as int,
+    EVector.fromBuffer(row[faceEmbeddingBlob] as List<int>).values,
+    row[faceScore] as double,
+    Detection.fromJson(json.decode(row[faceDetectionColumn] as String)),
+    row[faceBlur] as double,
+  );
+}

mobile/lib/face/model/box.dart

@@ -0,0 +1,42 @@
+/// Bounding box of a face. 
+/// 
+/// [`x`] and [y] are the coordinates of the top left corner of the box, so the minimim values
+/// [width] and [height] are the width and height of the box.
+/// All values are in absolute pixels relative to the original image size.
+class FaceBox {
+  final double x;
+  final double y;
+  final double width;
+  final double height;
+
+  FaceBox({
+    required this.x,
+    required this.y,
+    required this.width,
+    required this.height,
+  });
+
+  factory FaceBox.fromJson(Map<String, dynamic> json) {
+    return FaceBox(
+      x: (json['x'] is int
+          ? (json['x'] as int).toDouble()
+          : json['x'] as double),
+      y: (json['y'] is int
+          ? (json['y'] as int).toDouble()
+          : json['y'] as double),
+      width: (json['width'] is int
+          ? (json['width'] as int).toDouble()
+          : json['width'] as double),
+      height: (json['height'] is int
+          ? (json['height'] as int).toDouble()
+          : json['height'] as double),
+    );
+  }
+
+  Map<String, dynamic> toJson() => {
+        'x': x,
+        'y': y,
+        'width': width,
+        'height': height,
+      };
+}

mobile/lib/face/model/detection.dart

@@ -0,0 +1,37 @@
+import "package:photos/face/model/box.dart";
+import "package:photos/face/model/landmark.dart";
+
+class Detection {
+  FaceBox box;
+  List<Landmark> landmarks;
+
+  Detection({
+    required this.box,
+    required this.landmarks,
+  });
+
+  // emoty box
+  Detection.empty()
+      : box = FaceBox(
+          x: 0,
+          y: 0,
+          width: 0,
+          height: 0,
+        ),
+        landmarks = [];
+
+  Map<String, dynamic> toJson() => {
+        'box': box.toJson(),
+        'landmarks': landmarks.map((x) => x.toJson()).toList(),
+      };
+
+  factory Detection.fromJson(Map<String, dynamic> json) {
+    return Detection(
+      box: FaceBox.fromJson(json['box'] as Map<String, dynamic>),
+      landmarks: List<Landmark>.from(
+        json['landmarks']
+            .map((x) => Landmark.fromJson(x as Map<String, dynamic>)),
+      ),
+    );
+  }
+}

mobile/lib/face/model/face.dart

@@ -0,0 +1,43 @@
+import "package:photos/face/model/detection.dart";
+import "package:photos/services/face_ml/blur_detection/blur_constants.dart";
+
+class Face {
+  final int fileID;
+  final String faceID;
+  final List<double> embedding;
+  Detection detection;
+  final double score;
+  final double blur;
+
+  bool get isBlurry => blur < kLaplacianThreshold;
+
+  Face(
+    this.faceID,
+    this.fileID,
+    this.embedding,
+    this.score,
+    this.detection,
+    this.blur,
+  );
+
+  factory Face.fromJson(Map<String, dynamic> json) {
+    return Face(
+      json['faceID'] as String,
+      json['fileID'] as int,
+      List<double>.from(json['embeddings'] as List),
+      json['score'] as double,
+      Detection.fromJson(json['detection'] as Map<String, dynamic>),
+      // high value means t
+      (json['blur'] ?? kLapacianDefault) as double,
+    );
+  }
+
+  Map<String, dynamic> toJson() => {
+        'faceID': faceID,
+        'fileID': fileID,
+        'embeddings': embedding,
+        'detection': detection.toJson(),
+        'score': score,
+        'blur': blur,
+      };
+}

mobile/lib/face/model/landmark.dart

@@ -0,0 +1,26 @@
+// Class for the 'landmark' sub-object
+class Landmark {
+  double x;
+  double y;
+
+  Landmark({
+    required this.x,
+    required this.y,
+  });
+
+  Map<String, dynamic> toJson() => {
+        'x': x,
+        'y': y,
+      };
+
+  factory Landmark.fromJson(Map<String, dynamic> json) {
+    return Landmark(
+      x: (json['x'] is int
+          ? (json['x'] as int).toDouble()
+          : json['x'] as double),
+      y: (json['y'] is int
+          ? (json['y'] as int).toDouble()
+          : json['y'] as double),
+    );
+  }
+}

mobile/lib/face/model/person.dart

@@ -0,0 +1,70 @@
+class Person {
+  final String remoteID;
+  final PersonAttr attr;
+  Person(
+    this.remoteID,
+    this.attr,
+  );
+
+  // copyWith
+  Person copyWith({
+    String? remoteID,
+    PersonAttr? attr,
+  }) {
+    return Person(
+      remoteID ?? this.remoteID,
+      attr ?? this.attr,
+    );
+  }
+}
+
+class PersonAttr {
+  final String name;
+  final bool isHidden;
+  String? avatarFaceId;
+  final List<String> faces;
+  final String? birthDatae;
+  PersonAttr({
+    required this.name,
+    required this.faces,
+    this.avatarFaceId,
+    this.isHidden = false,
+    this.birthDatae,
+  });
+  // copyWith
+  PersonAttr copyWith({
+    String? name,
+    List<String>? faces,
+    String? avatarFaceId,
+    bool? isHidden,
+    String? birthDatae,
+  }) {
+    return PersonAttr(
+      name: name ?? this.name,
+      faces: faces ?? this.faces,
+      avatarFaceId: avatarFaceId ?? this.avatarFaceId,
+      isHidden: isHidden ?? this.isHidden,
+      birthDatae: birthDatae ?? this.birthDatae,
+    );
+  }
+
+  // toJson
+  Map<String, dynamic> toJson() => {
+        'name': name,
+        'faces': faces.toList(),
+        'avatarFaceId': avatarFaceId,
+        'isHidden': isHidden,
+        'birthDatae': birthDatae,
+      };
+
+  // fromJson
+  factory PersonAttr.fromJson(Map<String, dynamic> json) {
+    return PersonAttr(
+      name: json['name'] as String,
+      faces: List<String>.from(json['faces'] as List<dynamic>),
+      avatarFaceId: json['avatarFaceId'] as String?,
+      isHidden: json['isHidden'] as bool? ?? false,
+      birthDatae: json['birthDatae'] as String?,
+    );
+  }
+}

mobile/lib/face/model/person_face.dart

@@ -0,0 +1,37 @@
+import 'package:photos/face/model/face.dart';
+
+class PersonFace {
+  final Face face;
+  int? personID;
+  bool? confirmed;
+  double? closeDist;
+  String? closeFaceID;
+
+  PersonFace(
+    this.face,
+    this.personID,
+    this.closeDist,
+    this.closeFaceID, {
+    this.confirmed,
+  });
+
+  // toJson
+  Map<String, dynamic> toJson() => {
+        'face': face.toJson(),
+        'personID': personID,
+        'confirmed': confirmed ?? false,
+        'close_dist': closeDist,
+        'close_face_id': closeFaceID,
+      };
+
+  // fromJson
+  factory PersonFace.fromJson(Map<String, dynamic> json) {
+    return PersonFace(
+      Face.fromJson(json['face'] as Map<String, dynamic>),
+      json['personID'] as int?,
+      json['close_dist'] as double?,
+      json['close_face_id'] as String?,
+      confirmed: json['confirmed'] as bool?,
+    );
+  }
+}

mobile/lib/face/utils/import_from_zip.dart

@@ -0,0 +1,44 @@
+// import "dart:io";
+
+import "package:dio/dio.dart";
+import "package:logging/logging.dart";
+import "package:photos/core/configuration.dart";
+import "package:photos/core/network/network.dart";
+import "package:photos/face/model/face.dart";
+
+final _logger = Logger("import_from_zip");
+Future<List<Face>> downloadZip() async {
+  final List<Face> result = [];
+  for (int i = 0; i < 2; i++) {
+    _logger.info("downloading $i");
+    final remoteZipUrl = "http://192.168.1.13:8700/ml/cx_ml_json_${i}.json";
+    final response = await NetworkClient.instance.getDio().get(
+          remoteZipUrl,
+          options: Options(
+            headers: {"X-Auth-Token": Configuration.instance.getToken()},
+          ),
+        );
+
+    if (response.statusCode != 200) {
+      _logger.warning('download failed ${response.toString()}');
+      throw Exception("download failed");
+    }
+    final res = response.data as List<dynamic>;
+    for (final item in res) {
+      try {
+        result.add(Face.fromJson(item));
+      } catch (e) {
+        _logger.warning("failed to parse $item");
+        rethrow;
+      }
+    }
+  }
+  Set<String> faceID = {};
+  for (final face in result) {
+    if (faceID.contains(face.faceID)) {
+      _logger.warning("duplicate faceID ${face.faceID}");
+    }
+    faceID.add(face.faceID);
+  }
+  return result;
+}

mobile/lib/generated/intl/messages_en.dart

@@ -973,6 +973,7 @@ class MessageLookup extends MessageLookupByLibrary {
         "paymentFailedWithReason": m36,
         "pendingItems": MessageLookupByLibrary.simpleMessage("Pending items"),
         "pendingSync": MessageLookupByLibrary.simpleMessage("Pending sync"),
+        "people": MessageLookupByLibrary.simpleMessage("People"),
         "peopleUsingYourCode":
             MessageLookupByLibrary.simpleMessage("People using your code"),
         "permDeleteWarning": MessageLookupByLibrary.simpleMessage(

mobile/lib/generated/l10n.dart

@@ -8158,6 +8158,16 @@ class S {
     );
   }
 
+  /// `People`
+  String get people {
+    return Intl.message(
+      'People',
+      name: 'people',
+      desc: '',
+      args: [],
+    );
+  }
+
   /// `Contents`
   String get contents {
     return Intl.message(

mobile/lib/generated/protos/ente/common/box.pb.dart

@@ -0,0 +1,111 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/box.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:core' as $core;
+
+import 'package:protobuf/protobuf.dart' as $pb;
+
+/// CenterBox is a box where x,y is the center of the box
+class CenterBox extends $pb.GeneratedMessage {
+  factory CenterBox({
+    $core.double? x,
+    $core.double? y,
+    $core.double? height,
+    $core.double? width,
+  }) {
+    final $result = create();
+    if (x != null) {
+      $result.x = x;
+    }
+    if (y != null) {
+      $result.y = y;
+    }
+    if (height != null) {
+      $result.height = height;
+    }
+    if (width != null) {
+      $result.width = width;
+    }
+    return $result;
+  }
+  CenterBox._() : super();
+  factory CenterBox.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory CenterBox.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'CenterBox', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.common'), createEmptyInstance: create)
+    ..a<$core.double>(1, _omitFieldNames ? '' : 'x', $pb.PbFieldType.OF)
+    ..a<$core.double>(2, _omitFieldNames ? '' : 'y', $pb.PbFieldType.OF)
+    ..a<$core.double>(3, _omitFieldNames ? '' : 'height', $pb.PbFieldType.OF)
+    ..a<$core.double>(4, _omitFieldNames ? '' : 'width', $pb.PbFieldType.OF)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  CenterBox clone() => CenterBox()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  CenterBox copyWith(void Function(CenterBox) updates) => super.copyWith((message) => updates(message as CenterBox)) as CenterBox;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static CenterBox create() => CenterBox._();
+  CenterBox createEmptyInstance() => create();
+  static $pb.PbList<CenterBox> createRepeated() => $pb.PbList<CenterBox>();
+  @$core.pragma('dart2js:noInline')
+  static CenterBox getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<CenterBox>(create);
+  static CenterBox? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $core.double get x => $_getN(0);
+  @$pb.TagNumber(1)
+  set x($core.double v) { $_setFloat(0, v); }
+  @$pb.TagNumber(1)
+  $core.bool hasX() => $_has(0);
+  @$pb.TagNumber(1)
+  void clearX() => clearField(1);
+
+  @$pb.TagNumber(2)
+  $core.double get y => $_getN(1);
+  @$pb.TagNumber(2)
+  set y($core.double v) { $_setFloat(1, v); }
+  @$pb.TagNumber(2)
+  $core.bool hasY() => $_has(1);
+  @$pb.TagNumber(2)
+  void clearY() => clearField(2);
+
+  @$pb.TagNumber(3)
+  $core.double get height => $_getN(2);
+  @$pb.TagNumber(3)
+  set height($core.double v) { $_setFloat(2, v); }
+  @$pb.TagNumber(3)
+  $core.bool hasHeight() => $_has(2);
+  @$pb.TagNumber(3)
+  void clearHeight() => clearField(3);
+
+  @$pb.TagNumber(4)
+  $core.double get width => $_getN(3);
+  @$pb.TagNumber(4)
+  set width($core.double v) { $_setFloat(3, v); }
+  @$pb.TagNumber(4)
+  $core.bool hasWidth() => $_has(3);
+  @$pb.TagNumber(4)
+  void clearWidth() => clearField(4);
+}
+
+
+const _omitFieldNames = $core.bool.fromEnvironment('protobuf.omit_field_names');
+const _omitMessageNames = $core.bool.fromEnvironment('protobuf.omit_message_names');

mobile/lib/generated/protos/ente/common/box.pbenum.dart

@@ -0,0 +1,11 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/box.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+

mobile/lib/generated/protos/ente/common/box.pbjson.dart

@@ -0,0 +1,38 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/box.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:convert' as $convert;
+import 'dart:core' as $core;
+import 'dart:typed_data' as $typed_data;
+
+@$core.Deprecated('Use centerBoxDescriptor instead')
+const CenterBox$json = {
+  '1': 'CenterBox',
+  '2': [
+    {'1': 'x', '3': 1, '4': 1, '5': 2, '9': 0, '10': 'x', '17': true},
+    {'1': 'y', '3': 2, '4': 1, '5': 2, '9': 1, '10': 'y', '17': true},
+    {'1': 'height', '3': 3, '4': 1, '5': 2, '9': 2, '10': 'height', '17': true},
+    {'1': 'width', '3': 4, '4': 1, '5': 2, '9': 3, '10': 'width', '17': true},
+  ],
+  '8': [
+    {'1': '_x'},
+    {'1': '_y'},
+    {'1': '_height'},
+    {'1': '_width'},
+  ],
+};
+
+/// Descriptor for `CenterBox`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List centerBoxDescriptor = $convert.base64Decode(
+    'CglDZW50ZXJCb3gSEQoBeBgBIAEoAkgAUgF4iAEBEhEKAXkYAiABKAJIAVIBeYgBARIbCgZoZW'
+    'lnaHQYAyABKAJIAlIGaGVpZ2h0iAEBEhkKBXdpZHRoGAQgASgCSANSBXdpZHRoiAEBQgQKAl94'
+    'QgQKAl95QgkKB19oZWlnaHRCCAoGX3dpZHRo');
+

mobile/lib/generated/protos/ente/common/box.pbserver.dart

@@ -0,0 +1,14 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/box.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names
+// ignore_for_file: deprecated_member_use_from_same_package, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+export 'box.pb.dart';
+

mobile/lib/generated/protos/ente/common/point.pb.dart

@@ -0,0 +1,83 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/point.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:core' as $core;
+
+import 'package:protobuf/protobuf.dart' as $pb;
+
+/// EPoint is a point in 2D space
+class EPoint extends $pb.GeneratedMessage {
+  factory EPoint({
+    $core.double? x,
+    $core.double? y,
+  }) {
+    final $result = create();
+    if (x != null) {
+      $result.x = x;
+    }
+    if (y != null) {
+      $result.y = y;
+    }
+    return $result;
+  }
+  EPoint._() : super();
+  factory EPoint.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory EPoint.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'EPoint', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.common'), createEmptyInstance: create)
+    ..a<$core.double>(1, _omitFieldNames ? '' : 'x', $pb.PbFieldType.OF)
+    ..a<$core.double>(2, _omitFieldNames ? '' : 'y', $pb.PbFieldType.OF)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  EPoint clone() => EPoint()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  EPoint copyWith(void Function(EPoint) updates) => super.copyWith((message) => updates(message as EPoint)) as EPoint;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static EPoint create() => EPoint._();
+  EPoint createEmptyInstance() => create();
+  static $pb.PbList<EPoint> createRepeated() => $pb.PbList<EPoint>();
+  @$core.pragma('dart2js:noInline')
+  static EPoint getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<EPoint>(create);
+  static EPoint? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $core.double get x => $_getN(0);
+  @$pb.TagNumber(1)
+  set x($core.double v) { $_setFloat(0, v); }
+  @$pb.TagNumber(1)
+  $core.bool hasX() => $_has(0);
+  @$pb.TagNumber(1)
+  void clearX() => clearField(1);
+
+  @$pb.TagNumber(2)
+  $core.double get y => $_getN(1);
+  @$pb.TagNumber(2)
+  set y($core.double v) { $_setFloat(1, v); }
+  @$pb.TagNumber(2)
+  $core.bool hasY() => $_has(1);
+  @$pb.TagNumber(2)
+  void clearY() => clearField(2);
+}
+
+
+const _omitFieldNames = $core.bool.fromEnvironment('protobuf.omit_field_names');
+const _omitMessageNames = $core.bool.fromEnvironment('protobuf.omit_message_names');

mobile/lib/generated/protos/ente/common/point.pbenum.dart

@@ -0,0 +1,11 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/point.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+

mobile/lib/generated/protos/ente/common/point.pbjson.dart

@@ -0,0 +1,33 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/point.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:convert' as $convert;
+import 'dart:core' as $core;
+import 'dart:typed_data' as $typed_data;
+
+@$core.Deprecated('Use ePointDescriptor instead')
+const EPoint$json = {
+  '1': 'EPoint',
+  '2': [
+    {'1': 'x', '3': 1, '4': 1, '5': 2, '9': 0, '10': 'x', '17': true},
+    {'1': 'y', '3': 2, '4': 1, '5': 2, '9': 1, '10': 'y', '17': true},
+  ],
+  '8': [
+    {'1': '_x'},
+    {'1': '_y'},
+  ],
+};
+
+/// Descriptor for `EPoint`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List ePointDescriptor = $convert.base64Decode(
+    'CgZFUG9pbnQSEQoBeBgBIAEoAkgAUgF4iAEBEhEKAXkYAiABKAJIAVIBeYgBAUIECgJfeEIECg'
+    'JfeQ==');
+

mobile/lib/generated/protos/ente/common/point.pbserver.dart

@@ -0,0 +1,14 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/point.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names
+// ignore_for_file: deprecated_member_use_from_same_package, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+export 'point.pb.dart';
+

mobile/lib/generated/protos/ente/common/vector.pb.dart

@@ -0,0 +1,64 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/vector.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:core' as $core;
+
+import 'package:protobuf/protobuf.dart' as $pb;
+
+/// Vector is generic message for dealing with lists of doubles
+/// It should ideally be used independently and not as a submessage
+class EVector extends $pb.GeneratedMessage {
+  factory EVector({
+    $core.Iterable<$core.double>? values,
+  }) {
+    final $result = create();
+    if (values != null) {
+      $result.values.addAll(values);
+    }
+    return $result;
+  }
+  EVector._() : super();
+  factory EVector.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory EVector.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'EVector', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.common'), createEmptyInstance: create)
+    ..p<$core.double>(1, _omitFieldNames ? '' : 'values', $pb.PbFieldType.KD)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  EVector clone() => EVector()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  EVector copyWith(void Function(EVector) updates) => super.copyWith((message) => updates(message as EVector)) as EVector;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static EVector create() => EVector._();
+  EVector createEmptyInstance() => create();
+  static $pb.PbList<EVector> createRepeated() => $pb.PbList<EVector>();
+  @$core.pragma('dart2js:noInline')
+  static EVector getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<EVector>(create);
+  static EVector? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $core.List<$core.double> get values => $_getList(0);
+}
+
+
+const _omitFieldNames = $core.bool.fromEnvironment('protobuf.omit_field_names');
+const _omitMessageNames = $core.bool.fromEnvironment('protobuf.omit_message_names');

mobile/lib/generated/protos/ente/common/vector.pbenum.dart

@@ -0,0 +1,11 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/vector.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+

mobile/lib/generated/protos/ente/common/vector.pbjson.dart

@@ -0,0 +1,27 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/vector.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:convert' as $convert;
+import 'dart:core' as $core;
+import 'dart:typed_data' as $typed_data;
+
+@$core.Deprecated('Use eVectorDescriptor instead')
+const EVector$json = {
+  '1': 'EVector',
+  '2': [
+    {'1': 'values', '3': 1, '4': 3, '5': 1, '10': 'values'},
+  ],
+};
+
+/// Descriptor for `EVector`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List eVectorDescriptor = $convert.base64Decode(
+    'CgdFVmVjdG9yEhYKBnZhbHVlcxgBIAMoAVIGdmFsdWVz');
+

mobile/lib/generated/protos/ente/common/vector.pbserver.dart

@@ -0,0 +1,14 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/common/vector.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names
+// ignore_for_file: deprecated_member_use_from_same_package, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+export 'vector.pb.dart';
+

mobile/lib/generated/protos/ente/ml/face.pb.dart

@@ -0,0 +1,169 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/face.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:core' as $core;
+
+import 'package:protobuf/protobuf.dart' as $pb;
+
+import '../common/box.pb.dart' as $0;
+import '../common/point.pb.dart' as $1;
+
+class Detection extends $pb.GeneratedMessage {
+  factory Detection({
+    $0.CenterBox? box,
+    $1.EPoint? landmarks,
+  }) {
+    final $result = create();
+    if (box != null) {
+      $result.box = box;
+    }
+    if (landmarks != null) {
+      $result.landmarks = landmarks;
+    }
+    return $result;
+  }
+  Detection._() : super();
+  factory Detection.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory Detection.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'Detection', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.ml'), createEmptyInstance: create)
+    ..aOM<$0.CenterBox>(1, _omitFieldNames ? '' : 'box', subBuilder: $0.CenterBox.create)
+    ..aOM<$1.EPoint>(2, _omitFieldNames ? '' : 'landmarks', subBuilder: $1.EPoint.create)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  Detection clone() => Detection()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  Detection copyWith(void Function(Detection) updates) => super.copyWith((message) => updates(message as Detection)) as Detection;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static Detection create() => Detection._();
+  Detection createEmptyInstance() => create();
+  static $pb.PbList<Detection> createRepeated() => $pb.PbList<Detection>();
+  @$core.pragma('dart2js:noInline')
+  static Detection getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<Detection>(create);
+  static Detection? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $0.CenterBox get box => $_getN(0);
+  @$pb.TagNumber(1)
+  set box($0.CenterBox v) { setField(1, v); }
+  @$pb.TagNumber(1)
+  $core.bool hasBox() => $_has(0);
+  @$pb.TagNumber(1)
+  void clearBox() => clearField(1);
+  @$pb.TagNumber(1)
+  $0.CenterBox ensureBox() => $_ensure(0);
+
+  @$pb.TagNumber(2)
+  $1.EPoint get landmarks => $_getN(1);
+  @$pb.TagNumber(2)
+  set landmarks($1.EPoint v) { setField(2, v); }
+  @$pb.TagNumber(2)
+  $core.bool hasLandmarks() => $_has(1);
+  @$pb.TagNumber(2)
+  void clearLandmarks() => clearField(2);
+  @$pb.TagNumber(2)
+  $1.EPoint ensureLandmarks() => $_ensure(1);
+}
+
+class Face extends $pb.GeneratedMessage {
+  factory Face({
+    $core.String? id,
+    Detection? detection,
+    $core.double? confidence,
+  }) {
+    final $result = create();
+    if (id != null) {
+      $result.id = id;
+    }
+    if (detection != null) {
+      $result.detection = detection;
+    }
+    if (confidence != null) {
+      $result.confidence = confidence;
+    }
+    return $result;
+  }
+  Face._() : super();
+  factory Face.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory Face.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'Face', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.ml'), createEmptyInstance: create)
+    ..aOS(1, _omitFieldNames ? '' : 'id')
+    ..aOM<Detection>(2, _omitFieldNames ? '' : 'detection', subBuilder: Detection.create)
+    ..a<$core.double>(3, _omitFieldNames ? '' : 'confidence', $pb.PbFieldType.OF)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  Face clone() => Face()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  Face copyWith(void Function(Face) updates) => super.copyWith((message) => updates(message as Face)) as Face;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static Face create() => Face._();
+  Face createEmptyInstance() => create();
+  static $pb.PbList<Face> createRepeated() => $pb.PbList<Face>();
+  @$core.pragma('dart2js:noInline')
+  static Face getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<Face>(create);
+  static Face? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $core.String get id => $_getSZ(0);
+  @$pb.TagNumber(1)
+  set id($core.String v) { $_setString(0, v); }
+  @$pb.TagNumber(1)
+  $core.bool hasId() => $_has(0);
+  @$pb.TagNumber(1)
+  void clearId() => clearField(1);
+
+  @$pb.TagNumber(2)
+  Detection get detection => $_getN(1);
+  @$pb.TagNumber(2)
+  set detection(Detection v) { setField(2, v); }
+  @$pb.TagNumber(2)
+  $core.bool hasDetection() => $_has(1);
+  @$pb.TagNumber(2)
+  void clearDetection() => clearField(2);
+  @$pb.TagNumber(2)
+  Detection ensureDetection() => $_ensure(1);
+
+  @$pb.TagNumber(3)
+  $core.double get confidence => $_getN(2);
+  @$pb.TagNumber(3)
+  set confidence($core.double v) { $_setFloat(2, v); }
+  @$pb.TagNumber(3)
+  $core.bool hasConfidence() => $_has(2);
+  @$pb.TagNumber(3)
+  void clearConfidence() => clearField(3);
+}
+
+
+const _omitFieldNames = $core.bool.fromEnvironment('protobuf.omit_field_names');
+const _omitMessageNames = $core.bool.fromEnvironment('protobuf.omit_message_names');

mobile/lib/generated/protos/ente/ml/face.pbenum.dart

@@ -0,0 +1,11 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/face.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+

mobile/lib/generated/protos/ente/ml/face.pbjson.dart

@@ -0,0 +1,55 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/face.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:convert' as $convert;
+import 'dart:core' as $core;
+import 'dart:typed_data' as $typed_data;
+
+@$core.Deprecated('Use detectionDescriptor instead')
+const Detection$json = {
+  '1': 'Detection',
+  '2': [
+    {'1': 'box', '3': 1, '4': 1, '5': 11, '6': '.ente.common.CenterBox', '9': 0, '10': 'box', '17': true},
+    {'1': 'landmarks', '3': 2, '4': 1, '5': 11, '6': '.ente.common.EPoint', '9': 1, '10': 'landmarks', '17': true},
+  ],
+  '8': [
+    {'1': '_box'},
+    {'1': '_landmarks'},
+  ],
+};
+
+/// Descriptor for `Detection`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List detectionDescriptor = $convert.base64Decode(
+    'CglEZXRlY3Rpb24SLQoDYm94GAEgASgLMhYuZW50ZS5jb21tb24uQ2VudGVyQm94SABSA2JveI'
+    'gBARI2CglsYW5kbWFya3MYAiABKAsyEy5lbnRlLmNvbW1vbi5FUG9pbnRIAVIJbGFuZG1hcmtz'
+    'iAEBQgYKBF9ib3hCDAoKX2xhbmRtYXJrcw==');
+
+@$core.Deprecated('Use faceDescriptor instead')
+const Face$json = {
+  '1': 'Face',
+  '2': [
+    {'1': 'id', '3': 1, '4': 1, '5': 9, '9': 0, '10': 'id', '17': true},
+    {'1': 'detection', '3': 2, '4': 1, '5': 11, '6': '.ente.ml.Detection', '9': 1, '10': 'detection', '17': true},
+    {'1': 'confidence', '3': 3, '4': 1, '5': 2, '9': 2, '10': 'confidence', '17': true},
+  ],
+  '8': [
+    {'1': '_id'},
+    {'1': '_detection'},
+    {'1': '_confidence'},
+  ],
+};
+
+/// Descriptor for `Face`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List faceDescriptor = $convert.base64Decode(
+    'CgRGYWNlEhMKAmlkGAEgASgJSABSAmlkiAEBEjUKCWRldGVjdGlvbhgCIAEoCzISLmVudGUubW'
+    'wuRGV0ZWN0aW9uSAFSCWRldGVjdGlvbogBARIjCgpjb25maWRlbmNlGAMgASgCSAJSCmNvbmZp'
+    'ZGVuY2WIAQFCBQoDX2lkQgwKCl9kZXRlY3Rpb25CDQoLX2NvbmZpZGVuY2U=');
+

mobile/lib/generated/protos/ente/ml/face.pbserver.dart

@@ -0,0 +1,14 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/face.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names
+// ignore_for_file: deprecated_member_use_from_same_package, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+export 'face.pb.dart';
+

mobile/lib/generated/protos/ente/ml/fileml.pb.dart

@@ -0,0 +1,179 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/fileml.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:core' as $core;
+
+import 'package:fixnum/fixnum.dart' as $fixnum;
+import 'package:protobuf/protobuf.dart' as $pb;
+
+import 'face.pb.dart' as $2;
+
+class FileML extends $pb.GeneratedMessage {
+  factory FileML({
+    $fixnum.Int64? id,
+    $core.Iterable<$core.double>? clip,
+  }) {
+    final $result = create();
+    if (id != null) {
+      $result.id = id;
+    }
+    if (clip != null) {
+      $result.clip.addAll(clip);
+    }
+    return $result;
+  }
+  FileML._() : super();
+  factory FileML.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory FileML.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'FileML', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.ml'), createEmptyInstance: create)
+    ..aInt64(1, _omitFieldNames ? '' : 'id')
+    ..p<$core.double>(2, _omitFieldNames ? '' : 'clip', $pb.PbFieldType.KD)
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  FileML clone() => FileML()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  FileML copyWith(void Function(FileML) updates) => super.copyWith((message) => updates(message as FileML)) as FileML;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static FileML create() => FileML._();
+  FileML createEmptyInstance() => create();
+  static $pb.PbList<FileML> createRepeated() => $pb.PbList<FileML>();
+  @$core.pragma('dart2js:noInline')
+  static FileML getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<FileML>(create);
+  static FileML? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $fixnum.Int64 get id => $_getI64(0);
+  @$pb.TagNumber(1)
+  set id($fixnum.Int64 v) { $_setInt64(0, v); }
+  @$pb.TagNumber(1)
+  $core.bool hasId() => $_has(0);
+  @$pb.TagNumber(1)
+  void clearId() => clearField(1);
+
+  @$pb.TagNumber(2)
+  $core.List<$core.double> get clip => $_getList(1);
+}
+
+class FileFaces extends $pb.GeneratedMessage {
+  factory FileFaces({
+    $core.Iterable<$2.Face>? faces,
+    $core.int? height,
+    $core.int? width,
+    $core.int? version,
+    $core.String? error,
+  }) {
+    final $result = create();
+    if (faces != null) {
+      $result.faces.addAll(faces);
+    }
+    if (height != null) {
+      $result.height = height;
+    }
+    if (width != null) {
+      $result.width = width;
+    }
+    if (version != null) {
+      $result.version = version;
+    }
+    if (error != null) {
+      $result.error = error;
+    }
+    return $result;
+  }
+  FileFaces._() : super();
+  factory FileFaces.fromBuffer($core.List<$core.int> i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromBuffer(i, r);
+  factory FileFaces.fromJson($core.String i, [$pb.ExtensionRegistry r = $pb.ExtensionRegistry.EMPTY]) => create()..mergeFromJson(i, r);
+
+  static final $pb.BuilderInfo _i = $pb.BuilderInfo(_omitMessageNames ? '' : 'FileFaces', package: const $pb.PackageName(_omitMessageNames ? '' : 'ente.ml'), createEmptyInstance: create)
+    ..pc<$2.Face>(1, _omitFieldNames ? '' : 'faces', $pb.PbFieldType.PM, subBuilder: $2.Face.create)
+    ..a<$core.int>(2, _omitFieldNames ? '' : 'height', $pb.PbFieldType.O3)
+    ..a<$core.int>(3, _omitFieldNames ? '' : 'width', $pb.PbFieldType.O3)
+    ..a<$core.int>(4, _omitFieldNames ? '' : 'version', $pb.PbFieldType.O3)
+    ..aOS(5, _omitFieldNames ? '' : 'error')
+    ..hasRequiredFields = false
+  ;
+
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.deepCopy] instead. '
+  'Will be removed in next major version')
+  FileFaces clone() => FileFaces()..mergeFromMessage(this);
+  @$core.Deprecated(
+  'Using this can add significant overhead to your binary. '
+  'Use [GeneratedMessageGenericExtensions.rebuild] instead. '
+  'Will be removed in next major version')
+  FileFaces copyWith(void Function(FileFaces) updates) => super.copyWith((message) => updates(message as FileFaces)) as FileFaces;
+
+  $pb.BuilderInfo get info_ => _i;
+
+  @$core.pragma('dart2js:noInline')
+  static FileFaces create() => FileFaces._();
+  FileFaces createEmptyInstance() => create();
+  static $pb.PbList<FileFaces> createRepeated() => $pb.PbList<FileFaces>();
+  @$core.pragma('dart2js:noInline')
+  static FileFaces getDefault() => _defaultInstance ??= $pb.GeneratedMessage.$_defaultFor<FileFaces>(create);
+  static FileFaces? _defaultInstance;
+
+  @$pb.TagNumber(1)
+  $core.List<$2.Face> get faces => $_getList(0);
+
+  @$pb.TagNumber(2)
+  $core.int get height => $_getIZ(1);
+  @$pb.TagNumber(2)
+  set height($core.int v) { $_setSignedInt32(1, v); }
+  @$pb.TagNumber(2)
+  $core.bool hasHeight() => $_has(1);
+  @$pb.TagNumber(2)
+  void clearHeight() => clearField(2);
+
+  @$pb.TagNumber(3)
+  $core.int get width => $_getIZ(2);
+  @$pb.TagNumber(3)
+  set width($core.int v) { $_setSignedInt32(2, v); }
+  @$pb.TagNumber(3)
+  $core.bool hasWidth() => $_has(2);
+  @$pb.TagNumber(3)
+  void clearWidth() => clearField(3);
+
+  @$pb.TagNumber(4)
+  $core.int get version => $_getIZ(3);
+  @$pb.TagNumber(4)
+  set version($core.int v) { $_setSignedInt32(3, v); }
+  @$pb.TagNumber(4)
+  $core.bool hasVersion() => $_has(3);
+  @$pb.TagNumber(4)
+  void clearVersion() => clearField(4);
+
+  @$pb.TagNumber(5)
+  $core.String get error => $_getSZ(4);
+  @$pb.TagNumber(5)
+  set error($core.String v) { $_setString(4, v); }
+  @$pb.TagNumber(5)
+  $core.bool hasError() => $_has(4);
+  @$pb.TagNumber(5)
+  void clearError() => clearField(5);
+}
+
+
+const _omitFieldNames = $core.bool.fromEnvironment('protobuf.omit_field_names');
+const _omitMessageNames = $core.bool.fromEnvironment('protobuf.omit_message_names');

mobile/lib/generated/protos/ente/ml/fileml.pbenum.dart

@@ -0,0 +1,11 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/fileml.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+

mobile/lib/generated/protos/ente/ml/fileml.pbjson.dart

@@ -0,0 +1,57 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/fileml.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+import 'dart:convert' as $convert;
+import 'dart:core' as $core;
+import 'dart:typed_data' as $typed_data;
+
+@$core.Deprecated('Use fileMLDescriptor instead')
+const FileML$json = {
+  '1': 'FileML',
+  '2': [
+    {'1': 'id', '3': 1, '4': 1, '5': 3, '9': 0, '10': 'id', '17': true},
+    {'1': 'clip', '3': 2, '4': 3, '5': 1, '10': 'clip'},
+  ],
+  '8': [
+    {'1': '_id'},
+  ],
+};
+
+/// Descriptor for `FileML`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List fileMLDescriptor = $convert.base64Decode(
+    'CgZGaWxlTUwSEwoCaWQYASABKANIAFICaWSIAQESEgoEY2xpcBgCIAMoAVIEY2xpcEIFCgNfaW'
+    'Q=');
+
+@$core.Deprecated('Use fileFacesDescriptor instead')
+const FileFaces$json = {
+  '1': 'FileFaces',
+  '2': [
+    {'1': 'faces', '3': 1, '4': 3, '5': 11, '6': '.ente.ml.Face', '10': 'faces'},
+    {'1': 'height', '3': 2, '4': 1, '5': 5, '9': 0, '10': 'height', '17': true},
+    {'1': 'width', '3': 3, '4': 1, '5': 5, '9': 1, '10': 'width', '17': true},
+    {'1': 'version', '3': 4, '4': 1, '5': 5, '9': 2, '10': 'version', '17': true},
+    {'1': 'error', '3': 5, '4': 1, '5': 9, '9': 3, '10': 'error', '17': true},
+  ],
+  '8': [
+    {'1': '_height'},
+    {'1': '_width'},
+    {'1': '_version'},
+    {'1': '_error'},
+  ],
+};
+
+/// Descriptor for `FileFaces`. Decode as a `google.protobuf.DescriptorProto`.
+final $typed_data.Uint8List fileFacesDescriptor = $convert.base64Decode(
+    'CglGaWxlRmFjZXMSIwoFZmFjZXMYASADKAsyDS5lbnRlLm1sLkZhY2VSBWZhY2VzEhsKBmhlaW'
+    'dodBgCIAEoBUgAUgZoZWlnaHSIAQESGQoFd2lkdGgYAyABKAVIAVIFd2lkdGiIAQESHQoHdmVy'
+    'c2lvbhgEIAEoBUgCUgd2ZXJzaW9uiAEBEhkKBWVycm9yGAUgASgJSANSBWVycm9yiAEBQgkKB1'
+    '9oZWlnaHRCCAoGX3dpZHRoQgoKCF92ZXJzaW9uQggKBl9lcnJvcg==');
+

mobile/lib/generated/protos/ente/ml/fileml.pbserver.dart

@@ -0,0 +1,14 @@
+//
+//  Generated code. Do not modify.
+//  source: ente/ml/fileml.proto
+//
+// @dart = 2.12
+
+// ignore_for_file: annotate_overrides, camel_case_types, comment_references
+// ignore_for_file: constant_identifier_names
+// ignore_for_file: deprecated_member_use_from_same_package, library_prefixes
+// ignore_for_file: non_constant_identifier_names, prefer_final_fields
+// ignore_for_file: unnecessary_import, unnecessary_this, unused_import
+
+export 'fileml.pb.dart';
+

mobile/lib/l10n/intl_en.arb

@@ -1170,6 +1170,7 @@
     }
   },
   "faces": "Faces",
+  "people": "People",
   "contents": "Contents",
   "addNew": "Add new",
   "@addNew": {

mobile/lib/main.dart

@@ -25,6 +25,7 @@ import 'package:photos/services/app_lifecycle_service.dart';
 import 'package:photos/services/billing_service.dart';
 import 'package:photos/services/collections_service.dart';
 import "package:photos/services/entity_service.dart";
+import "package:photos/services/face_ml/face_ml_service.dart";
 import 'package:photos/services/favorites_service.dart';
 import 'package:photos/services/feature_flag_service.dart';
 import 'package:photos/services/local_file_update_service.dart';
@@ -242,9 +243,11 @@ Future<void> _init(bool isBackground, {String via = ''}) async {
   // Can not including existing tf/ml binaries as they are not being built
   // from source.
   // See https://gitlab.com/fdroid/fdroiddata/-/merge_requests/12671#note_1294346819
-  // if (!UpdateService.instance.isFdroidFlavor()) {
-  //   unawaited(ObjectDetectionService.instance.init());
-  // }
+  if (!UpdateService.instance.isFdroidFlavor()) {
+    // unawaited(ObjectDetectionService.instance.init());
+    unawaited(FaceMlService.instance.init());
+    FaceMlService.instance.listenIndexOnDiffSync();
+  }
 
   _logger.info("Initialization done");
 }

mobile/lib/models/gallery_type.dart

@@ -18,6 +18,8 @@ enum GalleryType {
   searchResults,
   locationTag,
   quickLink,
+  peopleTag,
+  cluster,
 }
 
 extension GalleyTypeExtension on GalleryType {
@@ -32,12 +34,14 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.locationTag:
       case GalleryType.quickLink:
       case GalleryType.uncategorized:
+      case GalleryType.peopleTag:
         return true;
 
       case GalleryType.hiddenSection:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.trash:
       case GalleryType.sharedCollection:
+      case GalleryType.cluster:
         return false;
     }
   }
@@ -50,6 +54,7 @@ extension GalleyTypeExtension on GalleryType {
         return true;
 
       case GalleryType.hiddenSection:
+      case GalleryType.peopleTag:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.favorite:
       case GalleryType.searchResults:
@@ -59,6 +64,7 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.trash:
       case GalleryType.sharedCollection:
       case GalleryType.locationTag:
+      case GalleryType.cluster:
         return false;
     }
   }
@@ -75,12 +81,14 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.uncategorized:
       case GalleryType.locationTag:
       case GalleryType.quickLink:
+      case GalleryType.peopleTag:
         return true;
       case GalleryType.trash:
       case GalleryType.archive:
       case GalleryType.hiddenSection:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.sharedCollection:
+      case GalleryType.cluster:
         return false;
     }
   }
@@ -98,8 +106,10 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.localFolder:
       case GalleryType.locationTag:
       case GalleryType.quickLink:
+      case GalleryType.peopleTag:
         return true;
       case GalleryType.trash:
+      case GalleryType.cluster:
       case GalleryType.sharedCollection:
         return false;
     }
@@ -114,8 +124,10 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.archive:
       case GalleryType.uncategorized:
       case GalleryType.locationTag:
+      case GalleryType.peopleTag:
         return true;
       case GalleryType.hiddenSection:
+      case GalleryType.cluster:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.localFolder:
       case GalleryType.trash:
@@ -132,6 +144,7 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.quickLink:
         return true;
       case GalleryType.hiddenSection:
+      case GalleryType.peopleTag:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.uncategorized:
       case GalleryType.favorite:
@@ -139,6 +152,7 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.homepage:
       case GalleryType.archive:
       case GalleryType.localFolder:
+      case GalleryType.cluster:
       case GalleryType.trash:
       case GalleryType.locationTag:
         return false;
@@ -154,6 +168,7 @@ extension GalleyTypeExtension on GalleryType {
         return true;
 
       case GalleryType.hiddenSection:
+      case GalleryType.peopleTag:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.favorite:
       case GalleryType.searchResults:
@@ -162,6 +177,7 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.trash:
       case GalleryType.sharedCollection:
       case GalleryType.locationTag:
+      case GalleryType.cluster:
         return false;
     }
   }
@@ -182,10 +198,12 @@ extension GalleyTypeExtension on GalleryType {
         return true;
 
       case GalleryType.hiddenSection:
+      case GalleryType.peopleTag:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.localFolder:
       case GalleryType.trash:
       case GalleryType.favorite:
+      case GalleryType.cluster:
       case GalleryType.sharedCollection:
         return false;
     }
@@ -203,12 +221,14 @@ extension GalleyTypeExtension on GalleryType {
       case GalleryType.searchResults:
       case GalleryType.uncategorized:
       case GalleryType.locationTag:
+      case GalleryType.peopleTag:
         return true;
 
       case GalleryType.hiddenSection:
       case GalleryType.hiddenOwnedCollection:
       case GalleryType.quickLink:
       case GalleryType.favorite:
+      case GalleryType.cluster:
       case GalleryType.archive:
       case GalleryType.localFolder:
       case GalleryType.trash:
@@ -244,7 +264,7 @@ extension GalleyTypeExtension on GalleryType {
   }
 
   bool showEditLocation() {
-    return this != GalleryType.sharedCollection;
+    return this != GalleryType.sharedCollection && this != GalleryType.cluster;
   }
 }
 
@@ -334,7 +354,9 @@ extension GalleryAppBarExtn on GalleryType {
       case GalleryType.locationTag:
       case GalleryType.searchResults:
         return false;
+      case GalleryType.cluster:
       case GalleryType.uncategorized:
+      case GalleryType.peopleTag:
       case GalleryType.ownedCollection:
       case GalleryType.sharedCollection:
       case GalleryType.quickLink:

mobile/lib/models/ml/ml_typedefs.dart

@@ -0,0 +1,7 @@
+typedef Embedding = List<double>;
+
+typedef Num3DInputMatrix = List<List<List<num>>>;
+
+typedef Int3DInputMatrix = List<List<List<int>>>;
+
+typedef Double3DInputMatrix = List<List<List<double>>>;

mobile/lib/models/ml/ml_versions.dart

@@ -0,0 +1,3 @@
+const faceMlVersion = 1;
+const clusterMlVersion = 1;
+const minimumClusterSize = 2;

mobile/lib/models/search/generic_search_result.dart

@@ -8,8 +8,15 @@ class GenericSearchResult extends SearchResult {
   final List<EnteFile> _files;
   final ResultType _type;
   final Function(BuildContext context)? onResultTap;
+  final Map<String, dynamic> params;
 
-  GenericSearchResult(this._type, this._name, this._files, {this.onResultTap});
+  GenericSearchResult(
+    this._type,
+    this._name,
+    this._files, {
+    this.onResultTap,
+    this.params = const {},
+  });
 
   @override
   String name() {

mobile/lib/models/search/search_constants.dart

@@ -0,0 +1,3 @@
+const kPersonParamID = 'person_id';
+const kClusterParamId = 'cluster_id';
+const kFileID = 'file_id';

mobile/lib/models/search/search_types.dart

@@ -33,6 +33,7 @@ enum ResultType {
   fileCaption,
   event,
   shared,
+  faces,
   magic,
 }
 
@@ -55,7 +56,7 @@ extension SectionTypeExtensions on SectionType {
   String sectionTitle(BuildContext context) {
     switch (this) {
       case SectionType.face:
-        return S.of(context).faces;
+        return S.of(context).people;
       case SectionType.content:
         return S.of(context).contents;
       case SectionType.moment:
@@ -99,7 +100,7 @@ extension SectionTypeExtensions on SectionType {
   bool get isCTAVisible {
     switch (this) {
       case SectionType.face:
-        return false;
+        return true;
       case SectionType.content:
         return false;
       case SectionType.moment:
@@ -117,6 +118,8 @@ extension SectionTypeExtensions on SectionType {
     }
   }
 
+  bool get sortByName => this != SectionType.face;
+
   bool get isEmptyCTAVisible {
     switch (this) {
       case SectionType.face:
@@ -245,8 +248,7 @@ extension SectionTypeExtensions on SectionType {
   }) {
     switch (this) {
       case SectionType.face:
-        return Future.value(List<GenericSearchResult>.empty());
-
+        return SearchService.instance.getAllFace(limit);
       case SectionType.content:
         return Future.value(List<GenericSearchResult>.empty());
 

mobile/lib/services/face_ml/blur_detection/blur_constants.dart

@@ -0,0 +1,2 @@
+const kLaplacianThreshold = 10;
+const kLapacianDefault = 10000.0;

mobile/lib/services/face_ml/blur_detection/blur_detection_service.dart

@@ -0,0 +1,115 @@
+import 'package:logging/logging.dart';
+import "package:photos/services/face_ml/blur_detection/blur_constants.dart";
+
+class BlurDetectionService {
+  final _logger = Logger('BlurDetectionService');
+
+  // singleton pattern
+  BlurDetectionService._privateConstructor();
+  static final instance = BlurDetectionService._privateConstructor();
+  factory BlurDetectionService() => instance;
+
+  Future<(bool, double)> predictIsBlurGrayLaplacian(
+    List<List<int>> grayImage, {
+    int threshold = kLaplacianThreshold,
+  }) async {
+    final List<List<int>> laplacian = _applyLaplacian(grayImage);
+    final double variance = _calculateVariance(laplacian);
+    _logger.info('Variance: $variance');
+    return (variance < threshold, variance);
+  }
+
+  double _calculateVariance(List<List<int>> matrix) {
+    final int numRows = matrix.length;
+    final int numCols = matrix[0].length;
+    final int totalElements = numRows * numCols;
+
+    // Calculate the mean
+    double mean = 0;
+    for (var row in matrix) {
+      for (var value in row) {
+        mean += value;
+      }
+    }
+    mean /= totalElements;
+
+    // Calculate the variance
+    double variance = 0;
+    for (var row in matrix) {
+      for (var value in row) {
+        final double diff = value - mean;
+        variance += diff * diff;
+      }
+    }
+    variance /= totalElements;
+
+    return variance;
+  }
+
+  List<List<int>> _padImage(List<List<int>> image) {
+    final int numRows = image.length;
+    final int numCols = image[0].length;
+
+    // Create a new matrix with extra padding
+    final List<List<int>> paddedImage = List.generate(
+      numRows + 2,
+      (i) => List.generate(numCols + 2, (j) => 0, growable: false),
+      growable: false,
+    );
+
+    // Copy original image into the center of the padded image
+    for (int i = 0; i < numRows; i++) {
+      for (int j = 0; j < numCols; j++) {
+        paddedImage[i + 1][j + 1] = image[i][j];
+      }
+    }
+
+    // Reflect padding
+    // Top and bottom rows
+    for (int j = 1; j <= numCols; j++) {
+      paddedImage[0][j] = paddedImage[2][j]; // Top row
+      paddedImage[numRows + 1][j] = paddedImage[numRows - 1][j]; // Bottom row
+    }
+    // Left and right columns
+    for (int i = 0; i < numRows + 2; i++) {
+      paddedImage[i][0] = paddedImage[i][2]; // Left column
+      paddedImage[i][numCols + 1] = paddedImage[i][numCols - 1]; // Right column
+    }
+
+    return paddedImage;
+  }
+
+  List<List<int>> _applyLaplacian(List<List<int>> image) {
+    final List<List<int>> paddedImage = _padImage(image);
+    final int numRows = image.length;
+    final int numCols = image[0].length;
+    final List<List<int>> outputImage = List.generate(
+      numRows,
+      (i) => List.generate(numCols, (j) => 0, growable: false),
+      growable: false,
+    );
+
+    // Define the Laplacian kernel
+    final List<List<int>> kernel = [
+      [0, 1, 0],
+      [1, -4, 1],
+      [0, 1, 0],
+    ];
+
+    // Apply the kernel to each pixel
+    for (int i = 0; i < numRows; i++) {
+      for (int j = 0; j < numCols; j++) {
+        int sum = 0;
+        for (int ki = 0; ki < 3; ki++) {
+          for (int kj = 0; kj < 3; kj++) {
+            sum += paddedImage[i + ki][j + kj] * kernel[ki][kj];
+          }
+        }
+        // Adjust the output value if necessary (e.g., clipping)
+        outputImage[i][j] = sum; //.clamp(0, 255);
+      }
+    }
+
+    return outputImage;
+  }
+}

mobile/lib/services/face_ml/face_alignment/alignment_result.dart

@@ -0,0 +1,36 @@
+class AlignmentResult {
+  final List<List<double>> affineMatrix; // 3x3
+  final List<double> center; // [x, y]
+  final double size; // 1 / scale
+  final double rotation; // atan2(simRotation[1][0], simRotation[0][0]);
+
+  AlignmentResult({required this.affineMatrix, required this.center, required this.size, required this.rotation});
+
+  AlignmentResult.empty()
+      : affineMatrix = <List<double>>[
+          [1, 0, 0],
+          [0, 1, 0],
+          [0, 0, 1],
+        ],
+        center = <double>[0, 0],
+        size = 1,
+        rotation = 0;
+
+  factory AlignmentResult.fromJson(Map<String, dynamic> json) {
+    return AlignmentResult(
+      affineMatrix: (json['affineMatrix'] as List)
+          .map((item) => List<double>.from(item))
+          .toList(),
+      center: List<double>.from(json['center'] as List),
+      size: json['size'] as double,
+      rotation: json['rotation'] as double,
+    );
+  }
+
+  Map<String, dynamic> toJson() => {
+        'affineMatrix': affineMatrix,
+        'center': center,
+        'size': size,
+        'rotation': rotation,
+      };
+}

mobile/lib/services/face_ml/face_alignment/similarity_transform.dart

@@ -0,0 +1,171 @@
+import 'dart:math' show atan2;
+import 'package:ml_linalg/linalg.dart';
+import 'package:photos/extensions/ml_linalg_extensions.dart';
+import "package:photos/services/face_ml/face_alignment/alignment_result.dart";
+
+/// Class to compute the similarity transform between two sets of points.
+///
+/// The class estimates the parameters of the similarity transformation via the `estimate` function.
+/// After estimation, the transformation can be applied to an image using the `warpAffine` function.
+class SimilarityTransform {
+  Matrix _params = Matrix.fromList([
+    [1.0, 0.0, 0.0],
+    [0.0, 1.0, 0.0],
+    [0, 0, 1],
+  ]);
+  List<double> _center = <double>[0, 0]; // [x, y]
+  double _size = 1; // 1 / scale
+  double _rotation = 0; // atan2(simRotation[1][0], simRotation[0][0]);
+
+  final arcface4Landmarks = [
+    <double>[38.2946, 51.6963],
+    <double>[73.5318, 51.5014],
+    <double>[56.0252, 71.7366],
+    <double>[56.1396, 92.2848],
+  ];
+  final arcface5Landmarks = [
+    <double>[38.2946, 51.6963],
+    <double>[73.5318, 51.5014],
+    <double>[56.0252, 71.7366],
+    <double>[41.5493, 92.3655],
+    <double>[70.7299, 92.2041],
+  ];
+  get arcfaceNormalized4 => arcface4Landmarks
+      .map((list) => list.map((value) => value / 112.0).toList())
+      .toList();
+  get arcfaceNormalized5 => arcface5Landmarks
+      .map((list) => list.map((value) => value / 112.0).toList())
+      .toList();
+
+  List<List<double>> get paramsList => _params.to2DList();
+
+  // singleton pattern
+  SimilarityTransform._privateConstructor();
+  static final instance = SimilarityTransform._privateConstructor();
+  factory SimilarityTransform() => instance;
+
+  void _cleanParams() {
+    _params = Matrix.fromList([
+      [1.0, 0.0, 0.0],
+      [0.0, 1.0, 0.0],
+      [0, 0, 1],
+    ]);
+    _center = <double>[0, 0];
+    _size = 1;
+    _rotation = 0;
+  }
+
+  /// Function to estimate the parameters of the affine transformation. These parameters are stored in the class variable params.
+  ///
+  /// Returns a tuple of (AlignmentResult, bool). The bool indicates whether the parameters are valid or not.
+  ///
+  /// Runs efficiently in about 1-3 ms after initial warm-up.
+  ///
+  /// It takes the source and destination points as input and returns the
+  /// parameters of the affine transformation as output. The function
+  /// returns false if the parameters cannot be estimated. The function
+  /// estimates the parameters by solving a least-squares problem using
+  /// the Umeyama algorithm, via [_umeyama].
+  (AlignmentResult, bool) estimate(List<List<double>> src) {
+    _cleanParams();
+    final (params, center, size, rotation) =
+        _umeyama(src, arcfaceNormalized5, true);
+    _params = params;
+    _center = center;
+    _size = size;
+    _rotation = rotation;
+    final alignmentResult = AlignmentResult(
+      affineMatrix: paramsList,
+      center: _center,
+      size: _size,
+      rotation: _rotation,
+    );
+    // We check for NaN in the transformation matrix params.
+    final isNoNanInParam =
+        !_params.asFlattenedList.any((element) => element.isNaN);
+    return (alignmentResult, isNoNanInParam);
+  }
+
+  static (Matrix, List<double>, double, double) _umeyama(
+    List<List<double>> src,
+    List<List<double>> dst, [
+    bool estimateScale = true,
+  ]) {
+    final srcMat = Matrix.fromList(
+      src,
+      // .map((list) => list.map((value) => value.toDouble()).toList())
+      // .toList(),
+    );
+    final dstMat = Matrix.fromList(dst);
+    final num = srcMat.rowCount;
+    final dim = srcMat.columnCount;
+
+    // Compute mean of src and dst.
+    final srcMean = srcMat.mean(Axis.columns);
+    final dstMean = dstMat.mean(Axis.columns);
+
+    // Subtract mean from src and dst.
+    final srcDemean = srcMat.mapRows((vector) => vector - srcMean);
+    final dstDemean = dstMat.mapRows((vector) => vector - dstMean);
+
+    // Eq. (38).
+    final A = (dstDemean.transpose() * srcDemean) / num;
+
+    // Eq. (39).
+    var d = Vector.filled(dim, 1.0);
+    if (A.determinant() < 0) {
+      d = d.set(dim - 1, -1);
+    }
+
+    var T = Matrix.identity(dim + 1);
+
+    final svdResult = A.svd();
+    final Matrix U = svdResult['U']!;
+    final Vector S = svdResult['S']!;
+    final Matrix V = svdResult['V']!;
+
+    // Eq. (40) and (43).
+    final rank = A.matrixRank();
+    if (rank == 0) {
+      return (T * double.nan, <double>[0, 0], 1, 0);
+    } else if (rank == dim - 1) {
+      if (U.determinant() * V.determinant() > 0) {
+        T = T.setSubMatrix(0, dim, 0, dim, U * V);
+      } else {
+        final s = d[dim - 1];
+        d = d.set(dim - 1, -1);
+        final replacement = U * Matrix.diagonal(d.toList()) * V;
+        T = T.setSubMatrix(0, dim, 0, dim, replacement);
+        d = d.set(dim - 1, s);
+      }
+    } else {
+      final replacement = U * Matrix.diagonal(d.toList()) * V;
+      T = T.setSubMatrix(0, dim, 0, dim, replacement);
+    }
+    final Matrix simRotation = U * Matrix.diagonal(d.toList()) * V;
+
+    var scale = 1.0;
+    if (estimateScale) {
+      // Eq. (41) and (42).
+      scale = 1.0 / srcDemean.variance(Axis.columns).sum() * (S * d).sum();
+    }
+
+    final subTIndices = Iterable<int>.generate(dim, (index) => index);
+    final subT = T.sample(rowIndices: subTIndices, columnIndices: subTIndices);
+    final newSubT = dstMean - (subT * srcMean) * scale;
+    T = T.setValues(0, dim, dim, dim + 1, newSubT);
+    final newNewSubT =
+        T.sample(rowIndices: subTIndices, columnIndices: subTIndices) * scale;
+    T = T.setSubMatrix(0, dim, 0, dim, newNewSubT);
+
+    // final List<double> translation = [T[0][2], T[1][2]];
+    // final simRotation = replacement?;
+    final size = 1 / scale;
+    final rotation = atan2(simRotation[1][0], simRotation[0][0]);
+    final meanTranslation = (dstMean - 0.5) * size;
+    final centerMat = srcMean - meanTranslation;
+    final List<double> center = [centerMat[0], centerMat[1]];
+
+    return (T, center, size, rotation);
+  }
+}

mobile/lib/services/face_ml/face_clustering/cosine_distance.dart

@@ -0,0 +1,55 @@
+import 'dart:math' show sqrt;
+
+/// Calculates the cosine distance between two embeddings/vectors.
+///
+/// Throws an ArgumentError if the vectors are of different lengths or
+/// if either of the vectors has a magnitude of zero.
+double cosineDistance(List<double> vector1, List<double> vector2) {
+  if (vector1.length != vector2.length) {
+    throw ArgumentError('Vectors must be the same length');
+  }
+
+  double dotProduct = 0.0;
+  double magnitude1 = 0.0;
+  double magnitude2 = 0.0;
+
+  for (int i = 0; i < vector1.length; i++) {
+    dotProduct += vector1[i] * vector2[i];
+    magnitude1 += vector1[i] * vector1[i];
+    magnitude2 += vector2[i] * vector2[i];
+  }
+
+  magnitude1 = sqrt(magnitude1);
+  magnitude2 = sqrt(magnitude2);
+
+  // Avoid division by zero. This should never happen. If it does, then one of the vectors contains only zeros.
+  if (magnitude1 == 0 || magnitude2 == 0) {
+    throw ArgumentError('Vectors must not have a magnitude of zero');
+  }
+
+  final double similarity = dotProduct / (magnitude1 * magnitude2);
+
+  // Cosine distance is the complement of cosine similarity
+  return 1.0 - similarity;
+}
+
+// cosineDistForNormVectors calculates the cosine distance between two normalized embeddings/vectors.
+@pragma('vm:entry-point')
+double cosineDistForNormVectors(List<double> vector1, List<double> vector2) {
+  if (vector1.length != vector2.length) {
+    throw ArgumentError('Vectors must be the same length');
+  }
+  double dotProduct = 0.0;
+  for (int i = 0; i < vector1.length; i++) {
+    dotProduct += vector1[i] * vector2[i];
+  }
+  return 1.0 - dotProduct;
+}
+
+double calculateSqrDistance(List<double> v1, List<double> v2) {
+  double sum = 0;
+  for (int i = 0; i < v1.length; i++) {
+    sum += (v1[i] - v2[i]) * (v1[i] - v2[i]);
+  }
+  return sqrt(sum);
+}

mobile/lib/services/face_ml/face_clustering/linear_clustering_service.dart

@@ -0,0 +1,405 @@
+import "dart:async";
+import "dart:developer";
+import "dart:isolate";
+import "dart:math" show max;
+import "dart:typed_data";
+
+import "package:logging/logging.dart";
+import "package:photos/generated/protos/ente/common/vector.pb.dart";
+import "package:photos/services/face_ml/face_clustering/cosine_distance.dart";
+import "package:synchronized/synchronized.dart";
+
+class FaceInfo {
+  final String faceID;
+  final List<double> embedding;
+  int? clusterId;
+  String? closestFaceId;
+  int? closestDist;
+  FaceInfo({
+    required this.faceID,
+    required this.embedding,
+    this.clusterId,
+  });
+}
+
+enum ClusterOperation { linearIncrementalClustering }
+
+class FaceLinearClustering {
+  final _logger = Logger("FaceLinearClustering");
+
+  Timer? _inactivityTimer;
+  final Duration _inactivityDuration = const Duration(seconds: 30);
+  int _activeTasks = 0;
+
+
+  final _initLock = Lock();
+
+  late Isolate _isolate;
+  late ReceivePort _receivePort = ReceivePort();
+  late SendPort _mainSendPort;
+
+  bool isSpawned = false;
+  bool isRunning = false;
+
+  static const recommendedDistanceThreshold = 0.3;
+
+  // singleton pattern
+  FaceLinearClustering._privateConstructor();
+
+  /// Use this instance to access the FaceClustering service.
+  /// e.g. `FaceLinearClustering.instance.predict(dataset)`
+  static final instance = FaceLinearClustering._privateConstructor();
+  factory FaceLinearClustering() => instance;
+
+  Future<void> init() async {
+    return _initLock.synchronized(() async {
+      if (isSpawned) return;
+
+      _receivePort = ReceivePort();
+
+      try {
+        _isolate = await Isolate.spawn(
+          _isolateMain,
+          _receivePort.sendPort,
+        );
+        _mainSendPort = await _receivePort.first as SendPort;
+        isSpawned = true;
+
+        _resetInactivityTimer();
+      } catch (e) {
+        _logger.severe('Could not spawn isolate', e);
+        isSpawned = false;
+      }
+    });
+  }
+
+  Future<void> ensureSpawned() async {
+    if (!isSpawned) {
+      await init();
+    }
+  }
+
+  /// The main execution function of the isolate.
+  static void _isolateMain(SendPort mainSendPort) async {
+    final receivePort = ReceivePort();
+    mainSendPort.send(receivePort.sendPort);
+
+    receivePort.listen((message) async {
+      final functionIndex = message[0] as int;
+      final function = ClusterOperation.values[functionIndex];
+      final args = message[1] as Map<String, dynamic>;
+      final sendPort = message[2] as SendPort;
+
+      try {
+        switch (function) {
+          case ClusterOperation.linearIncrementalClustering:
+            final input = args['input'] as Map<String, (int?, Uint8List)>;
+            final result = FaceLinearClustering._runLinearClustering(input);
+            sendPort.send(result);
+            break;
+        }
+      } catch (e, stackTrace) {
+        sendPort
+            .send({'error': e.toString(), 'stackTrace': stackTrace.toString()});
+      }
+    });
+  }
+
+  /// The common method to run any operation in the isolate. It sends the [message] to [_isolateMain] and waits for the result.
+  Future<dynamic> _runInIsolate(
+    (ClusterOperation, Map<String, dynamic>) message,
+  ) async {
+    await ensureSpawned();
+    _resetInactivityTimer();
+    final completer = Completer<dynamic>();
+    final answerPort = ReceivePort();
+
+    _activeTasks++;
+    _mainSendPort.send([message.$1.index, message.$2, answerPort.sendPort]);
+
+    answerPort.listen((receivedMessage) {
+      if (receivedMessage is Map && receivedMessage.containsKey('error')) {
+        // Handle the error
+        final errorMessage = receivedMessage['error'];
+        final errorStackTrace = receivedMessage['stackTrace'];
+        final exception = Exception(errorMessage);
+        final stackTrace = StackTrace.fromString(errorStackTrace);
+        completer.completeError(exception, stackTrace);
+      } else {
+        completer.complete(receivedMessage);
+      }
+    });
+    _activeTasks--;
+
+    return completer.future;
+  }
+
+  /// Resets a timer that kills the isolate after a certain amount of inactivity.
+  ///
+  /// Should be called after initialization (e.g. inside `init()`) and after every call to isolate (e.g. inside `_runInIsolate()`)
+  void _resetInactivityTimer() {
+    _inactivityTimer?.cancel();
+    _inactivityTimer = Timer(_inactivityDuration, () {
+      if (_activeTasks > 0) {
+        _logger.info('Tasks are still running. Delaying isolate disposal.');
+        // Optionally, reschedule the timer to check again later.
+        _resetInactivityTimer();
+      } else {
+        _logger.info(
+        'Clustering Isolate has been inactive for ${_inactivityDuration.inSeconds} seconds with no tasks running. Killing isolate.',
+      );
+        dispose();
+      }
+    });
+  }
+
+  /// Disposes the isolate worker.
+  void dispose() {
+    if (!isSpawned) return;
+
+    isSpawned = false;
+    _isolate.kill();
+    _receivePort.close();
+    _inactivityTimer?.cancel();
+  }
+
+  /// Runs the clustering algorithm on the given [input], in an isolate.
+  ///
+  /// Returns the clustering result, which is a list of clusters, where each cluster is a list of indices of the dataset.
+  ///
+  /// WARNING: Make sure to always input data in the same ordering, otherwise the clustering can less less deterministic.
+  Future<Map<String, int>?> predict(
+    Map<String, (int?, Uint8List)> input,
+  ) async {
+    if (input.isEmpty) {
+      _logger.warning(
+        "Clustering dataset of embeddings is empty, returning empty list.",
+      );
+      return null;
+    }
+    if (isRunning) {
+      _logger.warning("Clustering is already running, returning empty list.");
+      return null;
+    }
+
+    isRunning = true;
+
+    // Clustering inside the isolate
+    _logger.info(
+      "Start clustering on ${input.length} embeddings inside computer isolate",
+    );
+    final stopwatchClustering = Stopwatch()..start();
+    // final Map<String, int> faceIdToCluster =
+    //     await _runLinearClusteringInComputer(input);
+    final Map<String, int> faceIdToCluster = await _runInIsolate(
+      (ClusterOperation.linearIncrementalClustering, {'input': input}),
+    );
+    // return _runLinearClusteringInComputer(input);
+    _logger.info(
+      'Clustering executed in ${stopwatchClustering.elapsed.inSeconds} seconds',
+    );
+
+    isRunning = false;
+
+    return faceIdToCluster;
+  }
+
+  static Map<String, int> _runLinearClustering(
+    Map<String, (int?, Uint8List)> x,
+  ) {
+    log(
+      "[ClusterIsolate] ${DateTime.now()} Copied to isolate ${x.length} faces",
+    );
+    final List<FaceInfo> faceInfos = [];
+    for (final entry in x.entries) {
+      faceInfos.add(
+        FaceInfo(
+          faceID: entry.key,
+          embedding: EVector.fromBuffer(entry.value.$2).values,
+          clusterId: entry.value.$1,
+        ),
+      );
+    }
+    // Sort the faceInfos such that the ones with null clusterId are at the end
+    faceInfos.sort((a, b) {
+      if (a.clusterId == null && b.clusterId == null) {
+        return 0;
+      } else if (a.clusterId == null) {
+        return 1;
+      } else if (b.clusterId == null) {
+        return -1;
+      } else {
+        return 0;
+      }
+    });
+    // Count the amount of null values at the end
+    int nullCount = 0;
+    for (final faceInfo in faceInfos.reversed) {
+      if (faceInfo.clusterId == null) {
+        nullCount++;
+      } else {
+        break;
+      }
+    }
+    log(
+      "[ClusterIsolate] ${DateTime.now()} Clustering $nullCount new faces without clusterId, and ${faceInfos.length - nullCount} faces with clusterId",
+    );
+    for (final clusteredFaceInfo
+        in faceInfos.sublist(0, faceInfos.length - nullCount)) {
+      assert(clusteredFaceInfo.clusterId != null);
+    }
+
+    final int totalFaces = faceInfos.length;
+    int clusterID = 1;
+    if (faceInfos.isNotEmpty) {
+      faceInfos.first.clusterId = clusterID;
+    }
+    log(
+      "[ClusterIsolate] ${DateTime.now()} Processing $totalFaces faces",
+    );
+    final stopwatchClustering = Stopwatch()..start();
+    for (int i = 1; i < totalFaces; i++) {
+      // Incremental clustering, so we can skip faces that already have a clusterId
+      if (faceInfos[i].clusterId != null) {
+        clusterID = max(clusterID, faceInfos[i].clusterId!);
+        continue;
+      }
+      final currentEmbedding = faceInfos[i].embedding;
+      int closestIdx = -1;
+      double closestDistance = double.infinity;
+      if (i % 250 == 0) {
+        log("[ClusterIsolate] ${DateTime.now()} Processing $i faces");
+      }
+      for (int j = 0; j < i; j++) {
+        final double distance = cosineDistForNormVectors(
+          currentEmbedding,
+          faceInfos[j].embedding,
+        );
+        if (distance < closestDistance) {
+          closestDistance = distance;
+          closestIdx = j;
+        }
+      }
+
+      if (closestDistance < recommendedDistanceThreshold) {
+        if (faceInfos[closestIdx].clusterId == null) {
+          // Ideally this should never happen, but just in case log it
+          log(
+            " [ClusterIsolate] ${DateTime.now()} Found new cluster $clusterID",
+          );
+          clusterID++;
+          faceInfos[closestIdx].clusterId = clusterID;
+        }
+        faceInfos[i].clusterId = faceInfos[closestIdx].clusterId;
+      } else {
+        clusterID++;
+        faceInfos[i].clusterId = clusterID;
+      }
+    }
+    final Map<String, int> result = {};
+    for (final faceInfo in faceInfos) {
+      result[faceInfo.faceID] = faceInfo.clusterId!;
+    }
+    stopwatchClustering.stop();
+    log(
+      ' [ClusterIsolate] ${DateTime.now()} Clustering for ${faceInfos.length} embeddings (${faceInfos[0].embedding.length} size) executed in ${stopwatchClustering.elapsedMilliseconds}ms, clusters $clusterID',
+    );
+    // return result;
+
+    // NOTe: The main clustering logic is done, the following is just filtering and logging
+    final input = x;
+    final faceIdToCluster = result;
+    stopwatchClustering.reset();
+    stopwatchClustering.start();
+
+    final Set<String> newFaceIds = <String>{};
+    input.forEach((key, value) {
+      if (value.$1 == null) {
+        newFaceIds.add(key);
+      }
+    });
+
+    //  Find faceIDs that are part of a cluster which is larger than 5 and are new faceIDs
+    final Map<int, int> clusterIdToSize = {};
+    faceIdToCluster.forEach((key, value) {
+      if (clusterIdToSize.containsKey(value)) {
+        clusterIdToSize[value] = clusterIdToSize[value]! + 1;
+      } else {
+        clusterIdToSize[value] = 1;
+      }
+    });
+    final Map<String, int> faceIdToClusterFiltered = {};
+    for (final entry in faceIdToCluster.entries) {
+      if (clusterIdToSize[entry.value]! > 0 && newFaceIds.contains(entry.key)) {
+        faceIdToClusterFiltered[entry.key] = entry.value;
+      }
+    }
+
+    // print top 10 cluster ids and their sizes based on the internal cluster id
+    final clusterIds = faceIdToCluster.values.toSet();
+    final clusterSizes = clusterIds.map((clusterId) {
+      return faceIdToCluster.values.where((id) => id == clusterId).length;
+    }).toList();
+    clusterSizes.sort();
+    // find clusters whose size is graeter than 1
+    int oneClusterCount = 0;
+    int moreThan5Count = 0;
+    int moreThan10Count = 0;
+    int moreThan20Count = 0;
+    int moreThan50Count = 0;
+    int moreThan100Count = 0;
+
+    // for (int i = 0; i < clusterSizes.length; i++) {
+    //   if (clusterSizes[i] > 100) {
+    //     moreThan100Count++;
+    //   } else if (clusterSizes[i] > 50) {
+    //     moreThan50Count++;
+    //   } else if (clusterSizes[i] > 20) {
+    //     moreThan20Count++;
+    //   } else if (clusterSizes[i] > 10) {
+    //     moreThan10Count++;
+    //   } else if (clusterSizes[i] > 5) {
+    //     moreThan5Count++;
+    //   } else if (clusterSizes[i] == 1) {
+    //     oneClusterCount++;
+    //   }
+    // }
+    for (int i = 0; i < clusterSizes.length; i++) {
+      if (clusterSizes[i] > 100) {
+        moreThan100Count++;
+      }
+      if (clusterSizes[i] > 50) {
+        moreThan50Count++;
+      }
+      if (clusterSizes[i] > 20) {
+        moreThan20Count++;
+      }
+      if (clusterSizes[i] > 10) {
+        moreThan10Count++;
+      }
+      if (clusterSizes[i] > 5) {
+        moreThan5Count++;
+      }
+      if (clusterSizes[i] == 1) {
+        oneClusterCount++;
+      }
+    }
+    // print the metrics
+    log(
+      '[ClusterIsolate]  Total clusters ${clusterIds.length}, '
+      'oneClusterCount $oneClusterCount, '
+      'moreThan5Count $moreThan5Count, '
+      'moreThan10Count $moreThan10Count, '
+      'moreThan20Count $moreThan20Count, '
+      'moreThan50Count $moreThan50Count, '
+      'moreThan100Count $moreThan100Count',
+    );
+    stopwatchClustering.stop();
+    log(
+      "[ClusterIsolate]  Clustering additional steps took ${stopwatchClustering.elapsedMilliseconds} ms",
+    );
+
+    // log('Top clusters count ${clusterSizes.reversed.take(10).toList()}');
+    return faceIdToClusterFiltered;
+  }
+}

mobile/lib/services/face_ml/face_detection/detection.dart

@@ -0,0 +1,469 @@
+import 'dart:convert' show utf8;
+import 'dart:math' show sqrt, pow;
+import 'dart:ui' show Size;
+import 'package:crypto/crypto.dart' show sha256;
+
+abstract class Detection {
+  final double score;
+
+  Detection({required this.score});
+
+  const Detection.empty() : score = 0;
+
+  get width;
+  get height;
+
+  @override
+  String toString();
+}
+
+extension BBoxExtension on List<double> {
+  void roundBoxToDouble() {
+    final widthRounded = (this[2] - this[0]).roundToDouble();
+    final heightRounded = (this[3] - this[1]).roundToDouble();
+    this[0] = this[0].roundToDouble();
+    this[1] = this[1].roundToDouble();
+    this[2] = this[0] + widthRounded;
+    this[3] = this[1] + heightRounded;
+  }
+
+  // double get xMinBox =>
+  //     isNotEmpty ? this[0] : throw IndexError.withLength(0, length);
+  // double get yMinBox =>
+  //     length >= 2 ? this[1] : throw IndexError.withLength(1, length);
+  // double get xMaxBox =>
+  //     length >= 3 ? this[2] : throw IndexError.withLength(2, length);
+  // double get yMaxBox =>
+  //     length >= 4 ? this[3] : throw IndexError.withLength(3, length);
+}
+
+/// This class represents a face detection with relative coordinates in the range [0, 1].
+/// The coordinates are relative to the image size. The pattern for the coordinates is always [x, y], where x is the horizontal coordinate and y is the vertical coordinate.
+///
+/// The [score] attribute is a double representing the confidence of the face detection.
+///
+/// The [box] attribute is a list of 4 doubles, representing the coordinates of the bounding box of the face detection.
+/// The four values of the box in order are: [xMinBox, yMinBox, xMaxBox, yMaxBox].
+///
+/// The [allKeypoints] attribute is a list of 6 lists of 2 doubles, representing the coordinates of the keypoints of the face detection.
+/// The six lists of two values in order are: [leftEye, rightEye, nose, mouth, leftEar, rightEar]. Again, all in [x, y] order.
+class FaceDetectionRelative extends Detection {
+  final List<double> box;
+  final List<List<double>> allKeypoints;
+
+  double get xMinBox => box[0];
+  double get yMinBox => box[1];
+  double get xMaxBox => box[2];
+  double get yMaxBox => box[3];
+
+  List<double> get leftEye => allKeypoints[0];
+  List<double> get rightEye => allKeypoints[1];
+  List<double> get nose => allKeypoints[2];
+  List<double> get leftMouth => allKeypoints[3];
+  List<double> get rightMouth => allKeypoints[4];
+
+  FaceDetectionRelative({
+    required double score,
+    required List<double> box,
+    required List<List<double>> allKeypoints,
+  })  : assert(
+          box.every((e) => e >= -0.1 && e <= 1.1),
+          "Bounding box values must be in the range [0, 1], with only a small margin of error allowed.",
+        ),
+        assert(
+          allKeypoints
+              .every((sublist) => sublist.every((e) => e >= -0.1 && e <= 1.1)),
+          "All keypoints must be in the range [0, 1], with only a small margin of error allowed.",
+        ),
+        box = List<double>.from(box.map((e) => e.clamp(0.0, 1.0))),
+        allKeypoints = allKeypoints
+            .map(
+              (sublist) =>
+                  List<double>.from(sublist.map((e) => e.clamp(0.0, 1.0))),
+            )
+            .toList(),
+        super(score: score);
+
+  factory FaceDetectionRelative.zero() {
+    return FaceDetectionRelative(
+      score: 0,
+      box: <double>[0, 0, 0, 0],
+      allKeypoints: <List<double>>[
+        [0, 0],
+        [0, 0],
+        [0, 0],
+        [0, 0],
+        [0, 0],
+      ],
+    );
+  }
+
+  /// This is used to initialize the FaceDetectionRelative object with default values.
+  /// This constructor is useful because it can be used to initialize a FaceDetectionRelative object as a constant.
+  /// Contrary to the `FaceDetectionRelative.zero()` constructor, this one gives immutable attributes [box] and [allKeypoints].
+  FaceDetectionRelative.defaultInitialization()
+      : box = const <double>[0, 0, 0, 0],
+        allKeypoints = const <List<double>>[
+          [0, 0],
+          [0, 0],
+          [0, 0],
+          [0, 0],
+          [0, 0],
+        ],
+        super.empty();
+
+  FaceDetectionRelative getNearestDetection(
+    List<FaceDetectionRelative> detections,
+  ) {
+    if (detections.isEmpty) {
+      throw ArgumentError("The detection list cannot be empty.");
+    }
+
+    var nearestDetection = detections[0];
+    var minDistance = double.infinity;
+
+    // Calculate the center of the current instance
+    final centerX1 = (xMinBox + xMaxBox) / 2;
+    final centerY1 = (yMinBox + yMaxBox) / 2;
+
+    for (var detection in detections) {
+      final centerX2 = (detection.xMinBox + detection.xMaxBox) / 2;
+      final centerY2 = (detection.yMinBox + detection.yMaxBox) / 2;
+      final distance =
+          sqrt(pow(centerX2 - centerX1, 2) + pow(centerY2 - centerY1, 2));
+      if (distance < minDistance) {
+        minDistance = distance;
+        nearestDetection = detection;
+      }
+    }
+    return nearestDetection;
+  }
+
+  void transformRelativeToOriginalImage(
+    List<double> fromBox, // [xMin, yMin, xMax, yMax]
+    List<double> toBox, // [xMin, yMin, xMax, yMax]
+  ) {
+    // Return if all elements of fromBox and toBox are equal
+    for (int i = 0; i < fromBox.length; i++) {
+      if (fromBox[i] != toBox[i]) {
+        break;
+      }
+      if (i == fromBox.length - 1) {
+        return;
+      }
+    }
+
+    // Account for padding
+    final double paddingXRatio =
+        (fromBox[0] - toBox[0]) / (toBox[2] - toBox[0]);
+    final double paddingYRatio =
+        (fromBox[1] - toBox[1]) / (toBox[3] - toBox[1]);
+
+    // Calculate the scaling and translation
+    final double scaleX = (fromBox[2] - fromBox[0]) / (1 - 2 * paddingXRatio);
+    final double scaleY = (fromBox[3] - fromBox[1]) / (1 - 2 * paddingYRatio);
+    final double translateX = fromBox[0] - paddingXRatio * scaleX;
+    final double translateY = fromBox[1] - paddingYRatio * scaleY;
+
+    // Transform Box
+    _transformBox(box, scaleX, scaleY, translateX, translateY);
+
+    // Transform All Keypoints
+    for (int i = 0; i < allKeypoints.length; i++) {
+      allKeypoints[i] = _transformPoint(
+        allKeypoints[i],
+        scaleX,
+        scaleY,
+        translateX,
+        translateY,
+      );
+    }
+  }
+
+  void correctForMaintainedAspectRatio(
+    Size originalSize,
+    Size newSize,
+  ) {
+    // Return if both are the same size, meaning no scaling was done on both width and height
+    if (originalSize == newSize) {
+      return;
+    }
+
+    // Calculate the scaling
+    final double scaleX = originalSize.width / newSize.width;
+    final double scaleY = originalSize.height / newSize.height;
+    const double translateX = 0;
+    const double translateY = 0;
+
+    // Transform Box
+    _transformBox(box, scaleX, scaleY, translateX, translateY);
+
+    // Transform All Keypoints
+    for (int i = 0; i < allKeypoints.length; i++) {
+      allKeypoints[i] = _transformPoint(
+        allKeypoints[i],
+        scaleX,
+        scaleY,
+        translateX,
+        translateY,
+      );
+    }
+  }
+
+  void _transformBox(
+    List<double> box,
+    double scaleX,
+    double scaleY,
+    double translateX,
+    double translateY,
+  ) {
+    box[0] = (box[0] * scaleX + translateX).clamp(0.0, 1.0);
+    box[1] = (box[1] * scaleY + translateY).clamp(0.0, 1.0);
+    box[2] = (box[2] * scaleX + translateX).clamp(0.0, 1.0);
+    box[3] = (box[3] * scaleY + translateY).clamp(0.0, 1.0);
+  }
+
+  List<double> _transformPoint(
+    List<double> point,
+    double scaleX,
+    double scaleY,
+    double translateX,
+    double translateY,
+  ) {
+    return [
+      (point[0] * scaleX + translateX).clamp(0.0, 1.0),
+      (point[1] * scaleY + translateY).clamp(0.0, 1.0),
+    ];
+  }
+
+  FaceDetectionAbsolute toAbsolute({
+    required int imageWidth,
+    required int imageHeight,
+  }) {
+    final scoreCopy = score;
+    final boxCopy = List<double>.from(box, growable: false);
+    final allKeypointsCopy = allKeypoints
+        .map((sublist) => List<double>.from(sublist, growable: false))
+        .toList();
+
+    boxCopy[0] *= imageWidth;
+    boxCopy[1] *= imageHeight;
+    boxCopy[2] *= imageWidth;
+    boxCopy[3] *= imageHeight;
+    // final intbox = boxCopy.map((e) => e.toInt()).toList();
+
+    for (List<double> keypoint in allKeypointsCopy) {
+      keypoint[0] *= imageWidth;
+      keypoint[1] *= imageHeight;
+    }
+    // final intKeypoints =
+    //     allKeypointsCopy.map((e) => e.map((e) => e.toInt()).toList()).toList();
+    return FaceDetectionAbsolute(
+      score: scoreCopy,
+      box: boxCopy,
+      allKeypoints: allKeypointsCopy,
+    );
+  }
+
+  String toFaceID({required int fileID}) {
+    // Assert that the values are within the expected range
+    assert(
+      (xMinBox >= 0 && xMinBox <= 1) &&
+          (yMinBox >= 0 && yMinBox <= 1) &&
+          (xMaxBox >= 0 && xMaxBox <= 1) &&
+          (yMaxBox >= 0 && yMaxBox <= 1),
+      "Bounding box values must be in the range [0, 1]",
+    );
+
+    // Extract bounding box values
+    final String xMin =
+        xMinBox.clamp(0.0, 0.999999).toStringAsFixed(5).substring(2);
+    final String yMin =
+        yMinBox.clamp(0.0, 0.999999).toStringAsFixed(5).substring(2);
+    final String xMax =
+        xMaxBox.clamp(0.0, 0.999999).toStringAsFixed(5).substring(2);
+    final String yMax =
+        yMaxBox.clamp(0.0, 0.999999).toStringAsFixed(5).substring(2);
+
+    // Convert the bounding box values to string and concatenate
+    final String rawID = "${xMin}_${yMin}_${xMax}_$yMax";
+
+    // Hash the concatenated string using SHA256
+    final digest = sha256.convert(utf8.encode(rawID));
+
+    // Return the hexadecimal representation of the hash
+    return fileID.toString() + '_' + digest.toString();
+  }
+
+  /// This method is used to generate a faceID for a face detection that was manually added by the user.
+  static String toFaceIDEmpty({required int fileID}) {
+    return fileID.toString() + '_0';
+  }
+
+  /// This method is used to check if a faceID corresponds to a manually added face detection and not an actual face detection.
+  static bool isFaceIDEmpty(String faceID) {
+    return faceID.split('_')[1] == '0';
+  }
+
+  @override
+  String toString() {
+    return 'FaceDetectionRelative( with relative coordinates: \n score: $score \n Box: xMinBox: $xMinBox, yMinBox: $yMinBox, xMaxBox: $xMaxBox, yMaxBox: $yMaxBox, \n Keypoints: leftEye: $leftEye, rightEye: $rightEye, nose: $nose, leftMouth: $leftMouth, rightMouth: $rightMouth \n )';
+  }
+
+  Map<String, dynamic> toJson() {
+    return {
+      'score': score,
+      'box': box,
+      'allKeypoints': allKeypoints,
+    };
+  }
+
+  factory FaceDetectionRelative.fromJson(Map<String, dynamic> json) {
+    return FaceDetectionRelative(
+      score: (json['score'] as num).toDouble(),
+      box: List<double>.from(json['box']),
+      allKeypoints: (json['allKeypoints'] as List)
+          .map((item) => List<double>.from(item))
+          .toList(),
+    );
+  }
+
+  @override
+
+  /// The width of the bounding box of the face detection, in relative range [0, 1].
+  double get width => xMaxBox - xMinBox;
+  @override
+
+  /// The height of the bounding box of the face detection, in relative range [0, 1].
+  double get height => yMaxBox - yMinBox;
+}
+
+/// This class represents a face detection with absolute coordinates in pixels, in the range [0, imageWidth] for the horizontal coordinates and [0, imageHeight] for the vertical coordinates.
+/// The pattern for the coordinates is always [x, y], where x is the horizontal coordinate and y is the vertical coordinate.
+///
+/// The [score] attribute is a double representing the confidence of the face detection.
+///
+/// The [box] attribute is a list of 4 integers, representing the coordinates of the bounding box of the face detection.
+/// The four values of the box in order are: [xMinBox, yMinBox, xMaxBox, yMaxBox].
+///
+/// The [allKeypoints] attribute is a list of 6 lists of 2 integers, representing the coordinates of the keypoints of the face detection.
+/// The six lists of two values in order are: [leftEye, rightEye, nose, mouth, leftEar, rightEar]. Again, all in [x, y] order.
+class FaceDetectionAbsolute extends Detection {
+  final List<double> box;
+  final List<List<double>> allKeypoints;
+
+  double get xMinBox => box[0];
+  double get yMinBox => box[1];
+  double get xMaxBox => box[2];
+  double get yMaxBox => box[3];
+
+  List<double> get leftEye => allKeypoints[0];
+  List<double> get rightEye => allKeypoints[1];
+  List<double> get nose => allKeypoints[2];
+  List<double> get leftMouth => allKeypoints[3];
+  List<double> get rightMouth => allKeypoints[4];
+
+  FaceDetectionAbsolute({
+    required double score,
+    required this.box,
+    required this.allKeypoints,
+  }) : super(score: score);
+
+  factory FaceDetectionAbsolute._zero() {
+    return FaceDetectionAbsolute(
+      score: 0,
+      box: <double>[0, 0, 0, 0],
+      allKeypoints: <List<double>>[
+        [0, 0],
+        [0, 0],
+        [0, 0],
+        [0, 0],
+        [0, 0],
+      ],
+    );
+  }
+
+  FaceDetectionAbsolute.defaultInitialization()
+      : box = const <double>[0, 0, 0, 0],
+        allKeypoints = const <List<double>>[
+          [0, 0],
+          [0, 0],
+          [0, 0],
+          [0, 0],
+          [0, 0],
+        ],
+        super.empty();
+
+  @override
+  String toString() {
+    return 'FaceDetectionAbsolute( with absolute coordinates: \n score: $score \n Box: xMinBox: $xMinBox, yMinBox: $yMinBox, xMaxBox: $xMaxBox, yMaxBox: $yMaxBox, \n Keypoints: leftEye: $leftEye, rightEye: $rightEye, nose: $nose, leftMouth: $leftMouth, rightMouth: $rightMouth \n )';
+  }
+
+  Map<String, dynamic> toJson() {
+    return {
+      'score': score,
+      'box': box,
+      'allKeypoints': allKeypoints,
+    };
+  }
+
+  factory FaceDetectionAbsolute.fromJson(Map<String, dynamic> json) {
+    return FaceDetectionAbsolute(
+      score: (json['score'] as num).toDouble(),
+      box: List<double>.from(json['box']),
+      allKeypoints: (json['allKeypoints'] as List)
+          .map((item) => List<double>.from(item))
+          .toList(),
+    );
+  }
+
+  static FaceDetectionAbsolute empty = FaceDetectionAbsolute._zero();
+
+  @override
+
+  /// The width of the bounding box of the face detection, in number of pixels, range [0, imageWidth].
+  double get width => xMaxBox - xMinBox;
+  @override
+
+  /// The height of the bounding box of the face detection, in number of pixels, range [0, imageHeight].
+  double get height => yMaxBox - yMinBox;
+}
+
+List<FaceDetectionAbsolute> relativeToAbsoluteDetections({
+  required List<FaceDetectionRelative> relativeDetections,
+  required int imageWidth,
+  required int imageHeight,
+}) {
+  final numberOfDetections = relativeDetections.length;
+  final absoluteDetections = List<FaceDetectionAbsolute>.filled(
+    numberOfDetections,
+    FaceDetectionAbsolute._zero(),
+  );
+  for (var i = 0; i < relativeDetections.length; i++) {
+    final relativeDetection = relativeDetections[i];
+    final absoluteDetection = relativeDetection.toAbsolute(
+      imageWidth: imageWidth,
+      imageHeight: imageHeight,
+    );
+
+    absoluteDetections[i] = absoluteDetection;
+  }
+
+  return absoluteDetections;
+}
+
+/// Returns an enlarged version of the [box] by a factor of [factor].
+List<double> getEnlargedRelativeBox(List<double> box, [double factor = 2]) {
+  final boxCopy = List<double>.from(box, growable: false);
+  // The four values of the box in order are: [xMinBox, yMinBox, xMaxBox, yMaxBox].
+
+  final width = boxCopy[2] - boxCopy[0];
+  final height = boxCopy[3] - boxCopy[1];
+
+  boxCopy[0] -= width * (factor - 1) / 2;
+  boxCopy[1] -= height * (factor - 1) / 2;
+  boxCopy[2] += width * (factor - 1) / 2;
+  boxCopy[3] += height * (factor - 1) / 2;
+
+  return boxCopy;
+}

mobile/lib/services/face_ml/face_detection/naive_non_max_suppression.dart

@@ -0,0 +1,49 @@
+import 'dart:math' as math show max, min;
+
+import "package:photos/services/face_ml/face_detection/detection.dart";
+
+List<FaceDetectionRelative> naiveNonMaxSuppression({
+  required List<FaceDetectionRelative> detections,
+  required double iouThreshold,
+}) {
+  // Sort the detections by score, the highest first
+  detections.sort((a, b) => b.score.compareTo(a.score));
+
+  // Loop through the detections and calculate the IOU
+  for (var i = 0; i < detections.length - 1; i++) {
+    for (var j = i + 1; j < detections.length; j++) {
+      final iou = _calculateIOU(detections[i], detections[j]);
+      if (iou >= iouThreshold) {
+        detections.removeAt(j);
+        j--;
+      }
+    }
+  }
+  return detections;
+}
+
+double _calculateIOU(
+  FaceDetectionRelative detectionA,
+  FaceDetectionRelative detectionB,
+) {
+  final areaA = detectionA.width * detectionA.height;
+  final areaB = detectionB.width * detectionB.height;
+
+  final intersectionMinX = math.max(detectionA.xMinBox, detectionB.xMinBox);
+  final intersectionMinY = math.max(detectionA.yMinBox, detectionB.yMinBox);
+  final intersectionMaxX = math.min(detectionA.xMaxBox, detectionB.xMaxBox);
+  final intersectionMaxY = math.min(detectionA.yMaxBox, detectionB.yMaxBox);
+
+  final intersectionWidth = intersectionMaxX - intersectionMinX;
+  final intersectionHeight = intersectionMaxY - intersectionMinY;
+
+  if (intersectionWidth < 0 || intersectionHeight < 0) {
+    return 0.0; // If boxes do not overlap, IoU is 0
+  }
+
+  final intersectionArea = intersectionWidth * intersectionHeight;
+
+  final unionArea = areaA + areaB - intersectionArea;
+
+  return intersectionArea / unionArea;
+}

mobile/lib/services/face_ml/face_detection/yolov5face/onnx_face_detection.dart

@@ -0,0 +1,786 @@
+import "dart:async";
+import "dart:developer" as dev show log;
+import "dart:io" show File;
+import "dart:isolate";
+import 'dart:typed_data' show Float32List, Uint8List;
+
+import "package:computer/computer.dart";
+import 'package:flutter/material.dart';
+import 'package:logging/logging.dart';
+import 'package:onnxruntime/onnxruntime.dart';
+import "package:photos/services/face_ml/face_detection/detection.dart";
+import "package:photos/services/face_ml/face_detection/naive_non_max_suppression.dart";
+import "package:photos/services/face_ml/face_detection/yolov5face/yolo_face_detection_exceptions.dart";
+import "package:photos/services/face_ml/face_detection/yolov5face/yolo_filter_extract_detections.dart";
+import "package:photos/services/remote_assets_service.dart";
+import "package:photos/utils/image_ml_isolate.dart";
+import "package:photos/utils/image_ml_util.dart";
+import "package:synchronized/synchronized.dart";
+
+enum FaceDetectionOperation { yoloInferenceAndPostProcessing }
+
+class YoloOnnxFaceDetection {
+  static final _logger = Logger('YOLOFaceDetectionService');
+
+  final _computer = Computer.shared();
+
+  int sessionAddress = 0;
+
+  static const kModelBucketEndpoint = "https://models.ente.io/";
+  static const kRemoteBucketModelPath = "yolov5s_face_640_640_dynamic.onnx";
+  // static const kRemoteBucketModelPath = "yolov5n_face_640_640.onnx";
+  static const modelRemotePath = kModelBucketEndpoint + kRemoteBucketModelPath;
+
+  static const kInputWidth = 640;
+  static const kInputHeight = 640;
+  static const kIouThreshold = 0.4;
+  static const kMinScoreSigmoidThreshold = 0.8;
+
+  bool isInitialized = false;
+
+  // Isolate things
+  Timer? _inactivityTimer;
+  final Duration _inactivityDuration = const Duration(seconds: 30);
+
+  final _initLock = Lock();
+  final _computerLock = Lock();
+
+  late Isolate _isolate;
+  late ReceivePort _receivePort = ReceivePort();
+  late SendPort _mainSendPort;
+
+  bool isSpawned = false;
+  bool isRunning = false;
+
+  // singleton pattern
+  YoloOnnxFaceDetection._privateConstructor();
+
+  /// Use this instance to access the FaceDetection service. Make sure to call `init()` before using it.
+  /// e.g. `await FaceDetection.instance.init();`
+  ///
+  /// Then you can use `predict()` to get the bounding boxes of the faces, so `FaceDetection.instance.predict(imageData)`
+  ///
+  /// config options: yoloV5FaceN //
+  static final instance = YoloOnnxFaceDetection._privateConstructor();
+
+  factory YoloOnnxFaceDetection() => instance;
+
+  /// Check if the interpreter is initialized, if not initialize it with `loadModel()`
+  Future<void> init() async {
+    if (!isInitialized) {
+      _logger.info('init is called');
+      final model =
+          await RemoteAssetsService.instance.getAsset(modelRemotePath);
+      final startTime = DateTime.now();
+      // Doing this from main isolate since `rootBundle` cannot be accessed outside it
+      sessionAddress = await _computer.compute(
+        _loadModel,
+        param: {
+          "modelPath": model.path,
+        },
+      );
+      final endTime = DateTime.now();
+      _logger.info(
+        "Face detection model loaded, took: ${(endTime.millisecondsSinceEpoch - startTime.millisecondsSinceEpoch).toString()}ms",
+      );
+      if (sessionAddress != -1) {
+        isInitialized = true;
+      }
+    }
+  }
+
+  Future<void> release() async {
+    if (isInitialized) {
+      await _computer
+          .compute(_releaseModel, param: {'address': sessionAddress});
+      isInitialized = false;
+      sessionAddress = 0;
+    }
+  }
+
+  Future<void> initIsolate() async {
+    return _initLock.synchronized(() async {
+      if (isSpawned) return;
+
+      _receivePort = ReceivePort();
+
+      try {
+        _isolate = await Isolate.spawn(
+          _isolateMain,
+          _receivePort.sendPort,
+        );
+        _mainSendPort = await _receivePort.first as SendPort;
+        isSpawned = true;
+
+        _resetInactivityTimer();
+      } catch (e) {
+        _logger.severe('Could not spawn isolate', e);
+        isSpawned = false;
+      }
+    });
+  }
+
+  Future<void> ensureSpawnedIsolate() async {
+    if (!isSpawned) {
+      await initIsolate();
+    }
+  }
+
+  /// The main execution function of the isolate.
+  static void _isolateMain(SendPort mainSendPort) async {
+    final receivePort = ReceivePort();
+    mainSendPort.send(receivePort.sendPort);
+
+    receivePort.listen((message) async {
+      final functionIndex = message[0] as int;
+      final function = FaceDetectionOperation.values[functionIndex];
+      final args = message[1] as Map<String, dynamic>;
+      final sendPort = message[2] as SendPort;
+
+      try {
+        switch (function) {
+          case FaceDetectionOperation.yoloInferenceAndPostProcessing:
+            final inputImageList = args['inputImageList'] as Float32List;
+            final inputShape = args['inputShape'] as List<int>;
+            final newSize = args['newSize'] as Size;
+            final sessionAddress = args['sessionAddress'] as int;
+            final timeSentToIsolate = args['timeNow'] as DateTime;
+            final delaySentToIsolate =
+                DateTime.now().difference(timeSentToIsolate).inMilliseconds;
+
+            final Stopwatch stopwatchPrepare = Stopwatch()..start();
+            final inputOrt = OrtValueTensor.createTensorWithDataList(
+              inputImageList,
+              inputShape,
+            );
+            final inputs = {'input': inputOrt};
+            stopwatchPrepare.stop();
+            dev.log(
+              '[YOLOFaceDetectionService] data preparation is finished, in ${stopwatchPrepare.elapsedMilliseconds}ms',
+            );
+
+            stopwatchPrepare.reset();
+            stopwatchPrepare.start();
+            final runOptions = OrtRunOptions();
+            final session = OrtSession.fromAddress(sessionAddress);
+            stopwatchPrepare.stop();
+            dev.log(
+              '[YOLOFaceDetectionService] session preparation is finished, in ${stopwatchPrepare.elapsedMilliseconds}ms',
+            );
+
+            final stopwatchInterpreter = Stopwatch()..start();
+            late final List<OrtValue?> outputs;
+            try {
+              outputs = session.run(runOptions, inputs);
+            } catch (e, s) {
+              dev.log(
+                '[YOLOFaceDetectionService] Error while running inference: $e \n $s',
+              );
+              throw YOLOInterpreterRunException();
+            }
+            stopwatchInterpreter.stop();
+            dev.log(
+              '[YOLOFaceDetectionService] interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+            );
+
+            final relativeDetections =
+                _yoloPostProcessOutputs(outputs, newSize);
+
+            sendPort
+                .send((relativeDetections, delaySentToIsolate, DateTime.now()));
+            break;
+        }
+      } catch (e, stackTrace) {
+        sendPort
+            .send({'error': e.toString(), 'stackTrace': stackTrace.toString()});
+      }
+    });
+  }
+
+  /// The common method to run any operation in the isolate. It sends the [message] to [_isolateMain] and waits for the result.
+  Future<dynamic> _runInIsolate(
+    (FaceDetectionOperation, Map<String, dynamic>) message,
+  ) async {
+    await ensureSpawnedIsolate();
+    _resetInactivityTimer();
+    final completer = Completer<dynamic>();
+    final answerPort = ReceivePort();
+
+    _mainSendPort.send([message.$1.index, message.$2, answerPort.sendPort]);
+
+    answerPort.listen((receivedMessage) {
+      if (receivedMessage is Map && receivedMessage.containsKey('error')) {
+        // Handle the error
+        final errorMessage = receivedMessage['error'];
+        final errorStackTrace = receivedMessage['stackTrace'];
+        final exception = Exception(errorMessage);
+        final stackTrace = StackTrace.fromString(errorStackTrace);
+        completer.completeError(exception, stackTrace);
+      } else {
+        completer.complete(receivedMessage);
+      }
+    });
+
+    return completer.future;
+  }
+
+  /// Resets a timer that kills the isolate after a certain amount of inactivity.
+  ///
+  /// Should be called after initialization (e.g. inside `init()`) and after every call to isolate (e.g. inside `_runInIsolate()`)
+  void _resetInactivityTimer() {
+    _inactivityTimer?.cancel();
+    _inactivityTimer = Timer(_inactivityDuration, () {
+      _logger.info(
+        'Face detection (YOLO ONNX) Isolate has been inactive for ${_inactivityDuration.inSeconds} seconds. Killing isolate.',
+      );
+      disposeIsolate();
+    });
+  }
+
+  /// Disposes the isolate worker.
+  void disposeIsolate() {
+    if (!isSpawned) return;
+
+    isSpawned = false;
+    _isolate.kill();
+    _receivePort.close();
+    _inactivityTimer?.cancel();
+  }
+
+  /// Detects faces in the given image data.
+  Future<(List<FaceDetectionRelative>, Size)> predict(
+    Uint8List imageData,
+  ) async {
+    assert(isInitialized);
+
+    final stopwatch = Stopwatch()..start();
+
+    final stopwatchDecoding = Stopwatch()..start();
+    final (inputImageList, originalSize, newSize) =
+        await ImageMlIsolate.instance.preprocessImageYoloOnnx(
+      imageData,
+      normalize: true,
+      requiredWidth: kInputWidth,
+      requiredHeight: kInputHeight,
+      maintainAspectRatio: true,
+      quality: FilterQuality.medium,
+    );
+
+    // final input = [inputImageList];
+    final inputShape = [
+      1,
+      3,
+      kInputHeight,
+      kInputWidth,
+    ];
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      inputShape,
+    );
+    final inputs = {'input': inputOrt};
+    stopwatchDecoding.stop();
+    _logger.info(
+      'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+    );
+    _logger.info('original size: $originalSize \n new size: $newSize');
+
+    // Run inference
+    final stopwatchInterpreter = Stopwatch()..start();
+    List<OrtValue?>? outputs;
+    try {
+      final runOptions = OrtRunOptions();
+      final session = OrtSession.fromAddress(sessionAddress);
+      outputs = session.run(runOptions, inputs);
+      // inputOrt.release();
+      // runOptions.release();
+    } catch (e, s) {
+      _logger.severe('Error while running inference: $e \n $s');
+      throw YOLOInterpreterRunException();
+    }
+    stopwatchInterpreter.stop();
+    _logger.info(
+      'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+    );
+
+    final relativeDetections = _yoloPostProcessOutputs(outputs, newSize);
+
+    stopwatch.stop();
+    _logger.info(
+      'predict() face detection executed in ${stopwatch.elapsedMilliseconds}ms',
+    );
+
+    return (relativeDetections, originalSize);
+  }
+
+  /// Detects faces in the given image data.
+  static Future<(List<FaceDetectionRelative>, Size)> predictSync(
+    String imagePath,
+    int sessionAddress,
+  ) async {
+    assert(sessionAddress != 0 && sessionAddress != -1);
+
+    final stopwatch = Stopwatch()..start();
+
+    final stopwatchDecoding = Stopwatch()..start();
+    final imageData = await File(imagePath).readAsBytes();
+    final (inputImageList, originalSize, newSize) =
+        await preprocessImageToFloat32ChannelsFirst(
+      imageData,
+      normalization: 1,
+      requiredWidth: kInputWidth,
+      requiredHeight: kInputHeight,
+      maintainAspectRatio: true,
+      quality: FilterQuality.medium,
+    );
+
+    // final input = [inputImageList];
+    final inputShape = [
+      1,
+      3,
+      kInputHeight,
+      kInputWidth,
+    ];
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      inputShape,
+    );
+    final inputs = {'input': inputOrt};
+    stopwatchDecoding.stop();
+    dev.log(
+      'Face detection image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+    );
+    _logger.info(
+      'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+    );
+    _logger.info('original size: $originalSize \n new size: $newSize');
+
+    // Run inference
+    final stopwatchInterpreter = Stopwatch()..start();
+    List<OrtValue?>? outputs;
+    try {
+      final runOptions = OrtRunOptions();
+      final session = OrtSession.fromAddress(sessionAddress);
+      outputs = session.run(runOptions, inputs);
+      // inputOrt.release();
+      // runOptions.release();
+    } catch (e, s) {
+      _logger.severe('Error while running inference: $e \n $s');
+      throw YOLOInterpreterRunException();
+    }
+    stopwatchInterpreter.stop();
+    _logger.info(
+      'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+    );
+
+    final relativeDetections = _yoloPostProcessOutputs(outputs, newSize);
+
+    stopwatch.stop();
+    _logger.info(
+      'predict() face detection executed in ${stopwatch.elapsedMilliseconds}ms',
+    );
+
+    return (relativeDetections, originalSize);
+  }
+
+  /// Detects faces in the given image data.
+  Future<(List<FaceDetectionRelative>, Size)> predictInIsolate(
+    Uint8List imageData,
+  ) async {
+    await ensureSpawnedIsolate();
+    assert(isInitialized);
+
+    _logger.info('predictInIsolate() is called');
+
+    final stopwatch = Stopwatch()..start();
+
+    final stopwatchDecoding = Stopwatch()..start();
+    final (inputImageList, originalSize, newSize) =
+        await ImageMlIsolate.instance.preprocessImageYoloOnnx(
+      imageData,
+      normalize: true,
+      requiredWidth: kInputWidth,
+      requiredHeight: kInputHeight,
+      maintainAspectRatio: true,
+      quality: FilterQuality.medium,
+    );
+    // final input = [inputImageList];
+    final inputShape = [
+      1,
+      3,
+      kInputHeight,
+      kInputWidth,
+    ];
+
+    stopwatchDecoding.stop();
+    _logger.info(
+      'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+    );
+    _logger.info('original size: $originalSize \n new size: $newSize');
+
+    final (
+      List<FaceDetectionRelative> relativeDetections,
+      delaySentToIsolate,
+      timeSentToMain
+    ) = await _runInIsolate(
+      (
+        FaceDetectionOperation.yoloInferenceAndPostProcessing,
+        {
+          'inputImageList': inputImageList,
+          'inputShape': inputShape,
+          'newSize': newSize,
+          'sessionAddress': sessionAddress,
+          'timeNow': DateTime.now(),
+        }
+      ),
+    ) as (List<FaceDetectionRelative>, int, DateTime);
+
+    final delaySentToMain =
+        DateTime.now().difference(timeSentToMain).inMilliseconds;
+
+    stopwatch.stop();
+    _logger.info(
+      'predictInIsolate() face detection executed in ${stopwatch.elapsedMilliseconds}ms, with ${delaySentToIsolate}ms delay sent to isolate, and ${delaySentToMain}ms delay sent to main, for a total of ${delaySentToIsolate + delaySentToMain}ms delay due to isolate',
+    );
+
+    return (relativeDetections, originalSize);
+  }
+
+  Future<(List<FaceDetectionRelative>, Size)> predictInComputer(
+    String imagePath,
+  ) async {
+    assert(isInitialized);
+
+    _logger.info('predictInComputer() is called');
+
+    final stopwatch = Stopwatch()..start();
+
+    final stopwatchDecoding = Stopwatch()..start();
+    final imageData = await File(imagePath).readAsBytes();
+    final (inputImageList, originalSize, newSize) =
+        await ImageMlIsolate.instance.preprocessImageYoloOnnx(
+      imageData,
+      normalize: true,
+      requiredWidth: kInputWidth,
+      requiredHeight: kInputHeight,
+      maintainAspectRatio: true,
+      quality: FilterQuality.medium,
+    );
+    // final input = [inputImageList];
+    return await _computerLock.synchronized(() async {
+      final inputShape = [
+        1,
+        3,
+        kInputHeight,
+        kInputWidth,
+      ];
+
+      stopwatchDecoding.stop();
+      _logger.info(
+        'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+      );
+      _logger.info('original size: $originalSize \n new size: $newSize');
+
+      final (
+        List<FaceDetectionRelative> relativeDetections,
+        delaySentToIsolate,
+        timeSentToMain
+      ) = await _computer.compute(
+        inferenceAndPostProcess,
+        param: {
+          'inputImageList': inputImageList,
+          'inputShape': inputShape,
+          'newSize': newSize,
+          'sessionAddress': sessionAddress,
+          'timeNow': DateTime.now(),
+        },
+      ) as (List<FaceDetectionRelative>, int, DateTime);
+
+      final delaySentToMain =
+          DateTime.now().difference(timeSentToMain).inMilliseconds;
+
+      stopwatch.stop();
+      _logger.info(
+        'predictInIsolate() face detection executed in ${stopwatch.elapsedMilliseconds}ms, with ${delaySentToIsolate}ms delay sent to isolate, and ${delaySentToMain}ms delay sent to main, for a total of ${delaySentToIsolate + delaySentToMain}ms delay due to isolate',
+      );
+
+      return (relativeDetections, originalSize);
+    });
+  }
+
+  /// Detects faces in the given image data.
+  /// This method is optimized for batch processing.
+  ///
+  /// `imageDataList`: The image data to analyze.
+  ///
+  /// WARNING: Currently this method only returns the detections for the first image in the batch.
+  /// Change the function to output all detection before actually using it in production.
+  Future<List<FaceDetectionRelative>> predictBatch(
+    List<Uint8List> imageDataList,
+  ) async {
+    assert(isInitialized);
+
+    final stopwatch = Stopwatch()..start();
+
+    final stopwatchDecoding = Stopwatch()..start();
+    final List<Float32List> inputImageDataLists = [];
+    final List<(Size, Size)> originalAndNewSizeList = [];
+    int concatenatedImageInputsLength = 0;
+    for (final imageData in imageDataList) {
+      final (inputImageList, originalSize, newSize) =
+          await ImageMlIsolate.instance.preprocessImageYoloOnnx(
+        imageData,
+        normalize: true,
+        requiredWidth: kInputWidth,
+        requiredHeight: kInputHeight,
+        maintainAspectRatio: true,
+        quality: FilterQuality.medium,
+      );
+      inputImageDataLists.add(inputImageList);
+      originalAndNewSizeList.add((originalSize, newSize));
+      concatenatedImageInputsLength += inputImageList.length;
+    }
+
+    final inputImageList = Float32List(concatenatedImageInputsLength);
+
+    int offset = 0;
+    for (int i = 0; i < inputImageDataLists.length; i++) {
+      final inputImageData = inputImageDataLists[i];
+      inputImageList.setRange(
+        offset,
+        offset + inputImageData.length,
+        inputImageData,
+      );
+      offset += inputImageData.length;
+    }
+
+    // final input = [inputImageList];
+    final inputShape = [
+      inputImageDataLists.length,
+      3,
+      kInputHeight,
+      kInputWidth,
+    ];
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      inputShape,
+    );
+    final inputs = {'input': inputOrt};
+    stopwatchDecoding.stop();
+    _logger.info(
+      'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+    );
+    // _logger.info('original size: $originalSize \n new size: $newSize');
+
+    _logger.info('interpreter.run is called');
+    // Run inference
+    final stopwatchInterpreter = Stopwatch()..start();
+    List<OrtValue?>? outputs;
+    try {
+      final runOptions = OrtRunOptions();
+      final session = OrtSession.fromAddress(sessionAddress);
+      outputs = session.run(runOptions, inputs);
+      inputOrt.release();
+      runOptions.release();
+    } catch (e, s) {
+      _logger.severe('Error while running inference: $e \n $s');
+      throw YOLOInterpreterRunException();
+    }
+    stopwatchInterpreter.stop();
+    _logger.info(
+      'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms, or ${stopwatchInterpreter.elapsedMilliseconds / inputImageDataLists.length} ms per image',
+    );
+
+    _logger.info('outputs: $outputs');
+
+    const int imageOutputToUse = 0;
+
+    // // Get output tensors
+    final nestedResults =
+        outputs[0]?.value as List<List<List<double>>>; // [b, 25200, 16]
+    final selectedResults = nestedResults[imageOutputToUse]; // [25200, 16]
+
+    // final rawScores = <double>[];
+    // for (final result in firstResults) {
+    //   rawScores.add(result[4]);
+    // }
+    // final rawScoresCopy = List<double>.from(rawScores);
+    // rawScoresCopy.sort();
+    // _logger.info('rawScores minimum: ${rawScoresCopy.first}');
+    // _logger.info('rawScores maximum: ${rawScoresCopy.last}');
+
+    var relativeDetections = yoloOnnxFilterExtractDetections(
+      kMinScoreSigmoidThreshold,
+      kInputWidth,
+      kInputHeight,
+      results: selectedResults,
+    );
+
+    // Release outputs
+    for (var element in outputs) {
+      element?.release();
+    }
+
+    // Account for the fact that the aspect ratio was maintained
+    for (final faceDetection in relativeDetections) {
+      faceDetection.correctForMaintainedAspectRatio(
+        Size(
+          kInputWidth.toDouble(),
+          kInputHeight.toDouble(),
+        ),
+        originalAndNewSizeList[imageOutputToUse].$2,
+      );
+    }
+
+    // Non-maximum suppression to remove duplicate detections
+    relativeDetections = naiveNonMaxSuppression(
+      detections: relativeDetections,
+      iouThreshold: kIouThreshold,
+    );
+
+    if (relativeDetections.isEmpty) {
+      _logger.info('No face detected');
+      return <FaceDetectionRelative>[];
+    }
+
+    stopwatch.stop();
+    _logger.info(
+      'predict() face detection executed in ${stopwatch.elapsedMilliseconds}ms',
+    );
+
+    return relativeDetections;
+  }
+
+  static List<FaceDetectionRelative> _yoloPostProcessOutputs(
+    List<OrtValue?>? outputs,
+    Size newSize,
+  ) {
+    // // Get output tensors
+    final nestedResults =
+        outputs?[0]?.value as List<List<List<double>>>; // [1, 25200, 16]
+    final firstResults = nestedResults[0]; // [25200, 16]
+
+    // final rawScores = <double>[];
+    // for (final result in firstResults) {
+    //   rawScores.add(result[4]);
+    // }
+    // final rawScoresCopy = List<double>.from(rawScores);
+    // rawScoresCopy.sort();
+    // _logger.info('rawScores minimum: ${rawScoresCopy.first}');
+    // _logger.info('rawScores maximum: ${rawScoresCopy.last}');
+
+    var relativeDetections = yoloOnnxFilterExtractDetections(
+      kMinScoreSigmoidThreshold,
+      kInputWidth,
+      kInputHeight,
+      results: firstResults,
+    );
+
+    // Release outputs
+    // outputs?.forEach((element) {
+    //   element?.release();
+    // });
+
+    // Account for the fact that the aspect ratio was maintained
+    for (final faceDetection in relativeDetections) {
+      faceDetection.correctForMaintainedAspectRatio(
+        Size(
+          kInputWidth.toDouble(),
+          kInputHeight.toDouble(),
+        ),
+        newSize,
+      );
+    }
+
+    // Non-maximum suppression to remove duplicate detections
+    relativeDetections = naiveNonMaxSuppression(
+      detections: relativeDetections,
+      iouThreshold: kIouThreshold,
+    );
+
+    dev.log(
+      '[YOLOFaceDetectionService] ${relativeDetections.length} faces detected',
+    );
+
+    return relativeDetections;
+  }
+
+  /// Initialize the interpreter by loading the model file.
+  static Future<int> _loadModel(Map args) async {
+    final sessionOptions = OrtSessionOptions()
+      ..setInterOpNumThreads(1)
+      ..setIntraOpNumThreads(1)
+      ..setSessionGraphOptimizationLevel(GraphOptimizationLevel.ortEnableAll);
+    try {
+      // _logger.info('Loading face embedding model');
+      final session =
+          OrtSession.fromFile(File(args["modelPath"]), sessionOptions);
+      // _logger.info('Face embedding model loaded');
+      return session.address;
+    } catch (e, _) {
+      // _logger.severe('Face embedding model not loaded', e, s);
+    }
+    return -1;
+  }
+
+  static Future<void> _releaseModel(Map args) async {
+    final address = args['address'] as int;
+    if (address == 0) {
+      return;
+    }
+    final session = OrtSession.fromAddress(address);
+    session.release();
+    return;
+  }
+
+  static Future<(List<FaceDetectionRelative>, int, DateTime)>
+      inferenceAndPostProcess(
+    Map args,
+  ) async {
+    final inputImageList = args['inputImageList'] as Float32List;
+    final inputShape = args['inputShape'] as List<int>;
+    final newSize = args['newSize'] as Size;
+    final sessionAddress = args['sessionAddress'] as int;
+    final timeSentToIsolate = args['timeNow'] as DateTime;
+    final delaySentToIsolate =
+        DateTime.now().difference(timeSentToIsolate).inMilliseconds;
+
+    final Stopwatch stopwatchPrepare = Stopwatch()..start();
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      inputShape,
+    );
+    final inputs = {'input': inputOrt};
+    stopwatchPrepare.stop();
+    dev.log(
+      '[YOLOFaceDetectionService] data preparation is finished, in ${stopwatchPrepare.elapsedMilliseconds}ms',
+    );
+
+    stopwatchPrepare.reset();
+    stopwatchPrepare.start();
+    final runOptions = OrtRunOptions();
+    final session = OrtSession.fromAddress(sessionAddress);
+    stopwatchPrepare.stop();
+    dev.log(
+      '[YOLOFaceDetectionService] session preparation is finished, in ${stopwatchPrepare.elapsedMilliseconds}ms',
+    );
+
+    final stopwatchInterpreter = Stopwatch()..start();
+    late final List<OrtValue?> outputs;
+    try {
+      outputs = session.run(runOptions, inputs);
+    } catch (e, s) {
+      dev.log(
+        '[YOLOFaceDetectionService] Error while running inference: $e \n $s',
+      );
+      throw YOLOInterpreterRunException();
+    }
+    stopwatchInterpreter.stop();
+    dev.log(
+      '[YOLOFaceDetectionService] interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds} ms',
+    );
+
+    final relativeDetections = _yoloPostProcessOutputs(outputs, newSize);
+
+    return (relativeDetections, delaySentToIsolate, DateTime.now());
+  }
+}

mobile/lib/services/face_ml/face_detection/yolov5face/yolo_face_detection_exceptions.dart

@@ -0,0 +1,3 @@
+class YOLOInterpreterInitializationException implements Exception {}
+
+class YOLOInterpreterRunException implements Exception {}

mobile/lib/services/face_ml/face_detection/yolov5face/yolo_face_detection_options.dart

@@ -0,0 +1,31 @@
+import 'dart:math' as math show log;
+
+class FaceDetectionOptionsYOLO {
+  final double minScoreSigmoidThreshold;
+  final double iouThreshold;
+  final int inputWidth;
+  final int inputHeight;
+  final int numCoords;
+  final int numKeypoints;
+  final int numValuesPerKeypoint;
+  final int maxNumFaces;
+  final double scoreClippingThresh;
+  final double inverseSigmoidMinScoreThreshold;
+  final bool useSigmoidScore;
+  final bool flipVertically;
+
+  FaceDetectionOptionsYOLO({
+    required this.minScoreSigmoidThreshold,
+    required this.iouThreshold,
+    required this.inputWidth,
+    required this.inputHeight,
+    this.numCoords = 14,
+    this.numKeypoints = 5,
+    this.numValuesPerKeypoint = 2,
+    this.maxNumFaces = 100,
+    this.scoreClippingThresh = 100.0,
+    this.useSigmoidScore = true,
+    this.flipVertically = false,
+  }) : inverseSigmoidMinScoreThreshold =
+            math.log(minScoreSigmoidThreshold / (1 - minScoreSigmoidThreshold));
+}

mobile/lib/services/face_ml/face_detection/yolov5face/yolo_filter_extract_detections.dart

@@ -0,0 +1,81 @@
+import "package:photos/services/face_ml/face_detection/detection.dart";
+
+List<FaceDetectionRelative> yoloOnnxFilterExtractDetections(
+  double minScoreSigmoidThreshold,
+  int inputWidth,
+  int inputHeight, {
+  required List<List<double>> results, // // [25200, 16]
+}) {
+  final outputDetections = <FaceDetectionRelative>[];
+  final output = <List<double>>[];
+
+  // Go through the raw output and check the scores
+  for (final result in results) {
+    // Filter out raw detections with low scores
+    if (result[4] < minScoreSigmoidThreshold) {
+      continue;
+    }
+
+    // Get the raw detection
+    final rawDetection = List<double>.from(result);
+
+    // Append the processed raw detection to the output
+    output.add(rawDetection);
+  }
+
+  for (final List<double> rawDetection in output) {
+    // Get absolute bounding box coordinates in format [xMin, yMin, xMax, yMax] https://github.com/deepcam-cn/yolov5-face/blob/eb23d18defe4a76cc06449a61cd51004c59d2697/utils/general.py#L216
+    final xMinAbs = rawDetection[0] - rawDetection[2] / 2;
+    final yMinAbs = rawDetection[1] - rawDetection[3] / 2;
+    final xMaxAbs = rawDetection[0] + rawDetection[2] / 2;
+    final yMaxAbs = rawDetection[1] + rawDetection[3] / 2;
+
+    // Get the relative bounding box coordinates in format [xMin, yMin, xMax, yMax]
+    final box = [
+      xMinAbs / inputWidth,
+      yMinAbs / inputHeight,
+      xMaxAbs / inputWidth,
+      yMaxAbs / inputHeight,
+    ];
+
+    // Get the keypoints coordinates in format [x, y]
+    final allKeypoints = <List<double>>[
+      [
+        rawDetection[5] / inputWidth,
+        rawDetection[6] / inputHeight,
+      ],
+      [
+        rawDetection[7] / inputWidth,
+        rawDetection[8] / inputHeight,
+      ],
+      [
+        rawDetection[9] / inputWidth,
+        rawDetection[10] / inputHeight,
+      ],
+      [
+        rawDetection[11] / inputWidth,
+        rawDetection[12] / inputHeight,
+      ],
+      [
+        rawDetection[13] / inputWidth,
+        rawDetection[14] / inputHeight,
+      ],
+    ];
+
+    // Get the score
+    final score =
+        rawDetection[4]; // Or should it be rawDetection[4]*rawDetection[15]?
+
+    // Create the relative detection
+    final detection = FaceDetectionRelative(
+      score: score,
+      box: box,
+      allKeypoints: allKeypoints,
+    );
+
+    // Append the relative detection to the output
+    outputDetections.add(detection);
+  }
+
+  return outputDetections;
+}

mobile/lib/services/face_ml/face_detection/yolov5face/yolo_model_config.dart

@@ -0,0 +1,22 @@
+import "package:photos/services/face_ml/face_detection/yolov5face/yolo_face_detection_options.dart";
+import "package:photos/services/face_ml/model_file.dart";
+
+class YOLOModelConfig {
+  final String modelPath;
+  final FaceDetectionOptionsYOLO faceOptions;
+
+  YOLOModelConfig({
+    required this.modelPath,
+    required this.faceOptions,
+  });
+}
+
+final YOLOModelConfig yoloV5FaceS640x640DynamicBatchonnx = YOLOModelConfig(
+  modelPath: ModelFile.yoloV5FaceS640x640DynamicBatchonnx,
+  faceOptions: FaceDetectionOptionsYOLO(
+    minScoreSigmoidThreshold: 0.8,
+    iouThreshold: 0.4,
+    inputWidth: 640,
+    inputHeight: 640,
+  ),
+);

mobile/lib/services/face_ml/face_embedding/face_embedding_exceptions.dart

@@ -0,0 +1,11 @@
+class MobileFaceNetInterpreterInitializationException implements Exception {}
+
+class MobileFaceNetImagePreprocessingException implements Exception {}
+
+class MobileFaceNetEmptyInput implements Exception {}
+
+class MobileFaceNetWrongInputSize implements Exception {}
+
+class MobileFaceNetWrongInputRange implements Exception {}
+
+class MobileFaceNetInterpreterRunException implements Exception {}

mobile/lib/services/face_ml/face_embedding/face_embedding_options.dart

@@ -0,0 +1,15 @@
+class FaceEmbeddingOptions {
+  final int inputWidth;
+  final int inputHeight;
+  final int embeddingLength;
+  final int numChannels;
+  final bool preWhiten;
+
+  FaceEmbeddingOptions({
+    required this.inputWidth,
+    required this.inputHeight,
+    this.embeddingLength = 192,
+    this.numChannels = 3,
+    this.preWhiten = false,
+  });
+}

mobile/lib/services/face_ml/face_embedding/face_embedding_service.dart

@@ -0,0 +1,279 @@
+import 'dart:io';
+import "dart:math" show min, max, sqrt;
+// import 'dart:math' as math show min, max;
+import 'dart:typed_data' show Uint8List;
+
+import "package:flutter/foundation.dart";
+import "package:logging/logging.dart";
+import 'package:photos/models/ml/ml_typedefs.dart';
+import "package:photos/services/face_ml/face_detection/detection.dart";
+import "package:photos/services/face_ml/face_embedding/face_embedding_exceptions.dart";
+import "package:photos/services/face_ml/face_embedding/face_embedding_options.dart";
+import "package:photos/services/face_ml/face_embedding/mobilefacenet_model_config.dart";
+import 'package:photos/utils/image_ml_isolate.dart';
+import 'package:photos/utils/image_ml_util.dart';
+import 'package:tflite_flutter/tflite_flutter.dart';
+
+/// This class is responsible for running the MobileFaceNet model, and can be accessed through the singleton `FaceEmbedding.instance`.
+class FaceEmbedding {
+  Interpreter? _interpreter;
+  IsolateInterpreter? _isolateInterpreter;
+  int get getAddress => _interpreter!.address;
+
+  final outputShapes = <List<int>>[];
+  final outputTypes = <TensorType>[];
+
+  final _logger = Logger("FaceEmbeddingService");
+
+  final MobileFaceNetModelConfig config;
+  final FaceEmbeddingOptions embeddingOptions;
+  // singleton pattern
+  FaceEmbedding._privateConstructor({required this.config})
+      : embeddingOptions = config.faceEmbeddingOptions;
+
+  /// Use this instance to access the FaceEmbedding service. Make sure to call `init()` before using it.
+  /// e.g. `await FaceEmbedding.instance.init();`
+  ///
+  /// Then you can use `predict()` to get the embedding of a face, so `FaceEmbedding.instance.predict(imageData)`
+  ///
+  /// config options: faceEmbeddingEnte
+  static final instance =
+      FaceEmbedding._privateConstructor(config: faceEmbeddingEnte);
+  factory FaceEmbedding() => instance;
+
+  /// Check if the interpreter is initialized, if not initialize it with `loadModel()`
+  Future<void> init() async {
+    if (_interpreter == null || _isolateInterpreter == null) {
+      await _loadModel();
+    }
+  }
+
+  Future<void> dispose() async {
+    _logger.info('dispose() is called');
+
+    try {
+      _interpreter?.close();
+      _interpreter = null;
+      await _isolateInterpreter?.close();
+      _isolateInterpreter = null;
+    } catch (e) {
+      _logger.severe('Error while closing interpreter: $e');
+      rethrow;
+    }
+  }
+
+  /// WARNING: This function only works for one face at a time. it's better to use [predict], which can handle both single and multiple faces.
+  Future<List<double>> predictSingle(
+    Uint8List imageData,
+    FaceDetectionRelative face,
+  ) async {
+    assert(_interpreter != null && _isolateInterpreter != null);
+
+    final stopwatch = Stopwatch()..start();
+
+    // Image decoding and preprocessing
+    List<List<List<List<num>>>> input;
+    List output;
+    try {
+      final stopwatchDecoding = Stopwatch()..start();
+      final (inputImageMatrix, _, _, _, _) =
+          await ImageMlIsolate.instance.preprocessMobileFaceNet(
+        imageData,
+        [face],
+      );
+      input = inputImageMatrix;
+      stopwatchDecoding.stop();
+      _logger.info(
+        'Image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+      );
+
+      output = createEmptyOutputMatrix(outputShapes[0]);
+    } catch (e) {
+      _logger.severe('Error while decoding and preprocessing image: $e');
+      throw MobileFaceNetImagePreprocessingException();
+    }
+
+    _logger.info('interpreter.run is called');
+    // Run inference
+    try {
+      await _isolateInterpreter!.run(input, output);
+      // _interpreter!.run(input, output);
+      // ignore: avoid_catches_without_on_clauses
+    } catch (e) {
+      _logger.severe('Error while running inference: $e');
+      throw MobileFaceNetInterpreterRunException();
+    }
+    _logger.info('interpreter.run is finished');
+
+    // Get output tensors
+    final embedding = output[0] as List<double>;
+
+    // Normalize the embedding
+    final norm = sqrt(embedding.map((e) => e * e).reduce((a, b) => a + b));
+    for (int i = 0; i < embedding.length; i++) {
+      embedding[i] /= norm;
+    }
+
+    stopwatch.stop();
+    _logger.info(
+      'predict() executed in ${stopwatch.elapsedMilliseconds}ms',
+    );
+
+    // _logger.info(
+    //   'results (only first few numbers): embedding ${embedding.sublist(0, 5)}',
+    // );
+    // _logger.info(
+    //   'Mean of embedding: ${embedding.reduce((a, b) => a + b) / embedding.length}',
+    // );
+    // _logger.info(
+    //   'Max of embedding: ${embedding.reduce(math.max)}',
+    // );
+    // _logger.info(
+    //   'Min of embedding: ${embedding.reduce(math.min)}',
+    // );
+
+    return embedding;
+  }
+
+  Future<List<List<double>>> predict(
+    List<Num3DInputMatrix> inputImageMatrix,
+  ) async {
+    assert(_interpreter != null && _isolateInterpreter != null);
+
+    final stopwatch = Stopwatch()..start();
+
+    _checkPreprocessedInput(inputImageMatrix); // [inputHeight, inputWidth, 3]
+    final input = [inputImageMatrix];
+    // await encodeAndSaveData(inputImageMatrix, 'input_mobilefacenet');
+
+    final output = <int, Object>{};
+    final outputShape = outputShapes[0];
+    outputShape[0] = inputImageMatrix.length;
+    output[0] = createEmptyOutputMatrix(outputShape);
+    // for (int i = 0; i < faces.length; i++) {
+    //   output[i] = createEmptyOutputMatrix(outputShapes[0]);
+    // }
+
+    _logger.info('interpreter.run is called');
+    // Run inference
+    final stopwatchInterpreter = Stopwatch()..start();
+    try {
+      await _isolateInterpreter!.runForMultipleInputs(input, output);
+      // _interpreter!.runForMultipleInputs(input, output);
+      // ignore: avoid_catches_without_on_clauses
+    } catch (e) {
+      _logger.severe('Error while running inference: $e');
+      throw MobileFaceNetInterpreterRunException();
+    }
+    stopwatchInterpreter.stop();
+    _logger.info(
+      'interpreter.run is finished, in ${stopwatchInterpreter.elapsedMilliseconds}ms',
+    );
+    // _logger.info('output: $output');
+
+    // Get output tensors
+    final embeddings = <List<double>>[];
+    final outerEmbedding = output[0]! as Iterable<dynamic>;
+    for (int i = 0; i < inputImageMatrix.length; i++) {
+      final embedding = List<double>.from(outerEmbedding.toList()[i]);
+      // _logger.info("The $i-th embedding: $embedding");
+      embeddings.add(embedding);
+    }
+    // await encodeAndSaveData(embeddings, 'output_mobilefacenet');
+
+    // Normalize the embedding
+    for (int i = 0; i < embeddings.length; i++) {
+      final embedding = embeddings[i];
+      final norm = sqrt(embedding.map((e) => e * e).reduce((a, b) => a + b));
+      for (int j = 0; j < embedding.length; j++) {
+        embedding[j] /= norm;
+      }
+    }
+
+    stopwatch.stop();
+    _logger.info(
+      'predictBatch() executed in ${stopwatch.elapsedMilliseconds}ms',
+    );
+
+    return embeddings;
+  }
+
+  Future<void> _loadModel() async {
+    _logger.info('loadModel is called');
+
+    try {
+      final interpreterOptions = InterpreterOptions();
+
+      // Android Delegates
+      // TODO: Make sure this works on both platforms: Android and iOS
+      if (Platform.isAndroid) {
+        // Use GPU Delegate (GPU). WARNING: It doesn't work on emulator
+        // if (!kDebugMode) {
+        //   interpreterOptions.addDelegate(GpuDelegateV2());
+        // }
+        // Use XNNPACK Delegate (CPU)
+        interpreterOptions.addDelegate(XNNPackDelegate());
+      }
+
+      // iOS Delegates
+      if (Platform.isIOS) {
+        // Use Metal Delegate (GPU)
+        interpreterOptions.addDelegate(GpuDelegate());
+      }
+
+      // Load model from assets
+      _interpreter ??= await Interpreter.fromAsset(
+        config.modelPath,
+        options: interpreterOptions,
+      );
+      _isolateInterpreter ??=
+          await IsolateInterpreter.create(address: _interpreter!.address);
+
+      _logger.info('Interpreter created from asset: ${config.modelPath}');
+
+      // Get tensor input shape [1, 112, 112, 3]
+      final inputTensors = _interpreter!.getInputTensors().first;
+      _logger.info('Input Tensors: $inputTensors');
+      // Get tensour output shape [1, 192]
+      final outputTensors = _interpreter!.getOutputTensors();
+      final outputTensor = outputTensors.first;
+      _logger.info('Output Tensors: $outputTensor');
+
+      for (var tensor in outputTensors) {
+        outputShapes.add(tensor.shape);
+        outputTypes.add(tensor.type);
+      }
+      _logger.info('outputShapes: $outputShapes');
+      _logger.info('loadModel is finished');
+      // ignore: avoid_catches_without_on_clauses
+    } catch (e) {
+      _logger.severe('Error while creating interpreter: $e');
+      throw MobileFaceNetInterpreterInitializationException();
+    }
+  }
+
+  void _checkPreprocessedInput(
+    List<Num3DInputMatrix> inputMatrix,
+  ) {
+    final embeddingOptions = config.faceEmbeddingOptions;
+
+    if (inputMatrix.isEmpty) {
+      // Check if the input is empty
+      throw MobileFaceNetEmptyInput();
+    }
+
+    // Check if the input is the correct size
+    if (inputMatrix[0].length != embeddingOptions.inputHeight ||
+        inputMatrix[0][0].length != embeddingOptions.inputWidth) {
+      throw MobileFaceNetWrongInputSize();
+    }
+
+    final flattened = inputMatrix[0].expand((i) => i).expand((i) => i);
+    final minValue = flattened.reduce(min);
+    final maxValue = flattened.reduce(max);
+
+    if (minValue < -1 || maxValue > 1) {
+      throw MobileFaceNetWrongInputRange();
+    }
+  }
+}

mobile/lib/services/face_ml/face_embedding/mobilefacenet_model_config.dart

@@ -0,0 +1,20 @@
+import "package:photos/services/face_ml/face_embedding/face_embedding_options.dart";
+import "package:photos/services/face_ml/model_file.dart";
+
+class MobileFaceNetModelConfig {
+  final String modelPath;
+  final FaceEmbeddingOptions faceEmbeddingOptions;
+
+  MobileFaceNetModelConfig({
+    required this.modelPath,
+    required this.faceEmbeddingOptions,
+  });
+}
+
+final MobileFaceNetModelConfig faceEmbeddingEnte = MobileFaceNetModelConfig(
+  modelPath: ModelFile.faceEmbeddingEnte,
+  faceEmbeddingOptions: FaceEmbeddingOptions(
+    inputWidth: 112,
+    inputHeight: 112,
+  ),
+);

mobile/lib/services/face_ml/face_embedding/onnx_face_embedding.dart

@@ -0,0 +1,245 @@
+import "dart:io" show File;
+import 'dart:math' as math show max, min, sqrt;
+import 'dart:typed_data' show Float32List;
+
+import 'package:computer/computer.dart';
+import 'package:logging/logging.dart';
+import 'package:onnxruntime/onnxruntime.dart';
+import "package:photos/services/face_ml/face_detection/detection.dart";
+import "package:photos/services/remote_assets_service.dart";
+import "package:photos/utils/image_ml_isolate.dart";
+import "package:synchronized/synchronized.dart";
+
+class FaceEmbeddingOnnx {
+  static const kModelBucketEndpoint = "https://models.ente.io/";
+  static const kRemoteBucketModelPath = "mobilefacenet_opset15.onnx";
+  static const modelRemotePath = kModelBucketEndpoint + kRemoteBucketModelPath;
+
+  static const int kInputSize = 112;
+  static const int kEmbeddingSize = 192;
+
+  static final _logger = Logger('FaceEmbeddingOnnx');
+
+  bool isInitialized = false;
+  int sessionAddress = 0;
+
+  final _computer = Computer.shared();
+
+  final _computerLock = Lock();
+
+  // singleton pattern
+  FaceEmbeddingOnnx._privateConstructor();
+
+  /// Use this instance to access the FaceEmbedding service. Make sure to call `init()` before using it.
+  /// e.g. `await FaceEmbedding.instance.init();`
+  ///
+  /// Then you can use `predict()` to get the embedding of a face, so `FaceEmbedding.instance.predict(imageData)`
+  ///
+  /// config options: faceEmbeddingEnte
+  static final instance = FaceEmbeddingOnnx._privateConstructor();
+  factory FaceEmbeddingOnnx() => instance;
+
+  /// Check if the interpreter is initialized, if not initialize it with `loadModel()`
+  Future<void> init() async {
+    if (!isInitialized) {
+      _logger.info('init is called');
+      final model =
+          await RemoteAssetsService.instance.getAsset(modelRemotePath);
+      final startTime = DateTime.now();
+      // Doing this from main isolate since `rootBundle` cannot be accessed outside it
+      sessionAddress = await _computer.compute(
+        _loadModel,
+        param: {
+          "modelPath": model.path,
+        },
+      );
+      final endTime = DateTime.now();
+      _logger.info(
+        "Face embedding model loaded, took: ${(endTime.millisecondsSinceEpoch - startTime.millisecondsSinceEpoch).toString()}ms",
+      );
+      if (sessionAddress != -1) {
+        isInitialized = true;
+      }
+    }
+  }
+
+  Future<void> release() async {
+    if (isInitialized) {
+      await _computer.compute(_releaseModel, param: {'address': sessionAddress});
+      isInitialized = false;
+      sessionAddress = 0;
+    }
+  }
+
+  static Future<int> _loadModel(Map args) async {
+    final sessionOptions = OrtSessionOptions()
+      ..setInterOpNumThreads(1)
+      ..setIntraOpNumThreads(1)
+      ..setSessionGraphOptimizationLevel(GraphOptimizationLevel.ortEnableAll);
+    try {
+      // _logger.info('Loading face embedding model');
+      final session =
+          OrtSession.fromFile(File(args["modelPath"]), sessionOptions);
+      // _logger.info('Face embedding model loaded');
+      return session.address;
+    } catch (e, _) {
+      // _logger.severe('Face embedding model not loaded', e, s);
+    }
+    return -1;
+  }
+
+  static Future<void> _releaseModel(Map args) async {
+    final address = args['address'] as int;
+    if (address == 0) {
+      return;
+    }
+    final session = OrtSession.fromAddress(address);
+    session.release();
+    return;
+  }
+
+  Future<(List<double>, bool, double)> predictFromImageDataInComputer(
+    String imagePath,
+    FaceDetectionRelative face,
+  ) async {
+    assert(sessionAddress != 0 && sessionAddress != -1 && isInitialized);
+
+    try {
+      final stopwatchDecoding = Stopwatch()..start();
+      final (inputImageList, alignmentResults, isBlur, blurValue, _) =
+          await ImageMlIsolate.instance.preprocessMobileFaceNetOnnx(
+        imagePath,
+        [face],
+      );
+      stopwatchDecoding.stop();
+      _logger.info(
+        'MobileFaceNet image decoding and preprocessing is finished, in ${stopwatchDecoding.elapsedMilliseconds}ms',
+      );
+
+      final stopwatch = Stopwatch()..start();
+      _logger.info('MobileFaceNet interpreter.run is called');
+      final embedding = await _computer.compute(
+        inferFromMap,
+        param: {
+          'input': inputImageList,
+          'address': sessionAddress,
+          'inputSize': kInputSize,
+        },
+        taskName: 'createFaceEmbedding',
+      ) as List<double>;
+      stopwatch.stop();
+      _logger.info(
+        'MobileFaceNet interpreter.run is finished, in ${stopwatch.elapsedMilliseconds}ms',
+      );
+
+      _logger.info(
+        'MobileFaceNet results (only first few numbers): embedding ${embedding.sublist(0, 5)}',
+      );
+      _logger.info(
+        'Mean of embedding: ${embedding.reduce((a, b) => a + b) / embedding.length}',
+      );
+      _logger.info(
+        'Max of embedding: ${embedding.reduce(math.max)}',
+      );
+      _logger.info(
+        'Min of embedding: ${embedding.reduce(math.min)}',
+      );
+
+      return (embedding, isBlur[0], blurValue[0]);
+    } catch (e) {
+      _logger.info('MobileFaceNet Error while running inference: $e');
+      rethrow;
+    }
+  }
+
+  Future<List<List<double>>> predictInComputer(Float32List input) async {
+    assert(sessionAddress != 0 && sessionAddress != -1 && isInitialized);
+    return await _computerLock.synchronized(() async {
+      try {
+        final stopwatch = Stopwatch()..start();
+        _logger.info('MobileFaceNet interpreter.run is called');
+        final embeddings = await _computer.compute(
+          inferFromMap,
+          param: {
+            'input': input,
+            'address': sessionAddress,
+            'inputSize': kInputSize,
+          },
+          taskName: 'createFaceEmbedding',
+        ) as List<List<double>>;
+        stopwatch.stop();
+        _logger.info(
+          'MobileFaceNet interpreter.run is finished, in ${stopwatch.elapsedMilliseconds}ms',
+        );
+
+        return embeddings;
+      } catch (e) {
+        _logger.info('MobileFaceNet Error while running inference: $e');
+        rethrow;
+      }
+    });
+  }
+
+  static Future<List<List<double>>> predictSync(
+    Float32List input,
+    int sessionAddress,
+  ) async {
+    assert(sessionAddress != 0 && sessionAddress != -1);
+    try {
+      final stopwatch = Stopwatch()..start();
+      _logger.info('MobileFaceNet interpreter.run is called');
+      final embeddings = await infer(
+        input,
+        sessionAddress,
+        kInputSize,
+      );
+      stopwatch.stop();
+      _logger.info(
+        'MobileFaceNet interpreter.run is finished, in ${stopwatch.elapsedMilliseconds}ms',
+      );
+
+      return embeddings;
+    } catch (e) {
+      _logger.info('MobileFaceNet Error while running inference: $e');
+      rethrow;
+    }
+  }
+
+  static Future<List<List<double>>> inferFromMap(Map args) async {
+    final inputImageList = args['input'] as Float32List;
+    final address = args['address'] as int;
+    final inputSize = args['inputSize'] as int;
+    return await infer(inputImageList, address, inputSize);
+  }
+
+  static Future<List<List<double>>> infer(
+    Float32List inputImageList,
+    int address,
+    int inputSize,
+  ) async {
+    final runOptions = OrtRunOptions();
+    final int numberOfFaces =
+        inputImageList.length ~/ (inputSize * inputSize * 3);
+    final inputOrt = OrtValueTensor.createTensorWithDataList(
+      inputImageList,
+      [numberOfFaces, inputSize, inputSize, 3],
+    );
+    final inputs = {'img_inputs': inputOrt};
+    final session = OrtSession.fromAddress(address);
+    final List<OrtValue?> outputs = session.run(runOptions, inputs);
+    final embeddings = outputs[0]?.value as List<List<double>>;
+
+    for (final embedding in embeddings) {
+      double normalization = 0;
+      for (int i = 0; i < kEmbeddingSize; i++) {
+        normalization += embedding[i] * embedding[i];
+      }
+      final double sqrtNormalization = math.sqrt(normalization);
+      for (int i = 0; i < kEmbeddingSize; i++) {
+        embedding[i] = embedding[i] / sqrtNormalization;
+      }
+    }
+
+    return embeddings;
+  }
+}

mobile/lib/services/face_ml/face_feedback.dart/cluster_feedback.dart

@@ -0,0 +1,379 @@
+import "dart:convert";
+
+import "package:photos/services/face_ml/face_clustering/cosine_distance.dart";
+import "package:photos/services/face_ml/face_feedback.dart/feedback.dart";
+import "package:photos/services/face_ml/face_feedback.dart/feedback_types.dart";
+
+abstract class ClusterFeedback extends Feedback {
+  static final Map<FeedbackType, Function(String)> fromJsonStringRegistry = {
+    FeedbackType.deleteClusterFeedback: DeleteClusterFeedback.fromJsonString,
+    FeedbackType.mergeClusterFeedback: MergeClusterFeedback.fromJsonString,
+    FeedbackType.renameOrCustomThumbnailClusterFeedback:
+        RenameOrCustomThumbnailClusterFeedback.fromJsonString,
+    FeedbackType.removePhotosClusterFeedback:
+        RemovePhotosClusterFeedback.fromJsonString,
+    FeedbackType.addPhotosClusterFeedback:
+        AddPhotosClusterFeedback.fromJsonString,
+  };
+
+  final List<double> medoid;
+  final double medoidDistanceThreshold;
+  // TODO: work out the optimal distance threshold so there's never an overlap between clusters
+
+  ClusterFeedback(
+    FeedbackType type,
+    this.medoid,
+    this.medoidDistanceThreshold, {
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  }) : super(
+          type,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  /// Compares this feedback with another [ClusterFeedback] to see if they are similar enough that only one should be kept.
+  ///
+  /// It checks this by comparing the distance between the two medoids with the medoidDistanceThreshold of each feedback.
+  ///
+  /// Returns true if they are similar enough, false otherwise.
+  /// // TODO: Should it maybe return a merged feedback instead, when you are similar enough?
+  bool looselyMatchesMedoid(ClusterFeedback other) {
+    // Using the cosineDistance function you mentioned
+    final double distance = cosineDistance(medoid, other.medoid);
+
+    // Check if the distance is less than either of the threshold values
+    return distance < medoidDistanceThreshold ||
+        distance < other.medoidDistanceThreshold;
+  }
+
+  bool exactlyMatchesMedoid(ClusterFeedback other) {
+    if (medoid.length != other.medoid.length) {
+      return false;
+    }
+    for (int i = 0; i < medoid.length; i++) {
+      if (medoid[i] != other.medoid[i]) {
+        return false;
+      }
+    }
+    return true;
+  }
+}
+
+class DeleteClusterFeedback extends ClusterFeedback {
+  DeleteClusterFeedback({
+    required List<double> medoid,
+    required double medoidDistanceThreshold,
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  }) : super(
+          FeedbackType.deleteClusterFeedback,
+          medoid,
+          medoidDistanceThreshold,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  @override
+  Map<String, dynamic> toJson() {
+    return {
+      'type': type.toValueString(),
+      'medoid': medoid,
+      'medoidDistanceThreshold': medoidDistanceThreshold,
+      'feedbackID': feedbackID,
+      'timestamp': timestamp.toIso8601String(),
+      'madeOnFaceMlVersion': madeOnFaceMlVersion,
+      'madeOnClusterMlVersion': madeOnClusterMlVersion,
+    };
+  }
+
+  @override
+  String toJsonString() => jsonEncode(toJson());
+
+  static DeleteClusterFeedback fromJson(Map<String, dynamic> json) {
+    assert(json['type'] == FeedbackType.deleteClusterFeedback.toValueString());
+    return DeleteClusterFeedback(
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'],
+      feedbackID: json['feedbackID'],
+      timestamp: DateTime.parse(json['timestamp']),
+      madeOnFaceMlVersion: json['madeOnFaceMlVersion'],
+      madeOnClusterMlVersion: json['madeOnClusterMlVersion'],
+    );
+  }
+
+  static fromJsonString(String jsonString) {
+    return fromJson(jsonDecode(jsonString));
+  }
+}
+
+class MergeClusterFeedback extends ClusterFeedback {
+  final List<double> medoidToMoveTo;
+
+  MergeClusterFeedback({
+    required List<double> medoid,
+    required double medoidDistanceThreshold,
+    required this.medoidToMoveTo,
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  }) : super(
+          FeedbackType.mergeClusterFeedback,
+          medoid,
+          medoidDistanceThreshold,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  @override
+  Map<String, dynamic> toJson() {
+    return {
+      'type': type.toValueString(),
+      'medoid': medoid,
+      'medoidDistanceThreshold': medoidDistanceThreshold,
+      'medoidToMoveTo': medoidToMoveTo,
+      'feedbackID': feedbackID,
+      'timestamp': timestamp.toIso8601String(),
+      'madeOnFaceMlVersion': madeOnFaceMlVersion,
+      'madeOnClusterMlVersion': madeOnClusterMlVersion,
+    };
+  }
+
+  @override
+  String toJsonString() => jsonEncode(toJson());
+
+  static MergeClusterFeedback fromJson(Map<String, dynamic> json) {
+    assert(json['type'] == FeedbackType.mergeClusterFeedback.toValueString());
+    return MergeClusterFeedback(
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'],
+      medoidToMoveTo: (json['medoidToMoveTo'] as List?)
+              ?.map((item) => item as double)
+              .toList() ??
+          [],
+      feedbackID: json['feedbackID'],
+      timestamp: DateTime.parse(json['timestamp']),
+      madeOnFaceMlVersion: json['madeOnFaceMlVersion'],
+      madeOnClusterMlVersion: json['madeOnClusterMlVersion'],
+    );
+  }
+
+  static MergeClusterFeedback fromJsonString(String jsonString) {
+    return fromJson(jsonDecode(jsonString));
+  }
+}
+
+class RenameOrCustomThumbnailClusterFeedback extends ClusterFeedback {
+  String? customName;
+  String? customThumbnailFaceId;
+
+  RenameOrCustomThumbnailClusterFeedback({
+    required List<double> medoid,
+    required double medoidDistanceThreshold,
+    this.customName,
+    this.customThumbnailFaceId,
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  })  : assert(
+          customName != null || customThumbnailFaceId != null,
+          "Either customName or customThumbnailFaceId must be non-null!",
+        ),
+        super(
+          FeedbackType.renameOrCustomThumbnailClusterFeedback,
+          medoid,
+          medoidDistanceThreshold,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  @override
+  Map<String, dynamic> toJson() {
+    return {
+      'type': type.toValueString(),
+      'medoid': medoid,
+      'medoidDistanceThreshold': medoidDistanceThreshold,
+      if (customName != null) 'customName': customName,
+      if (customThumbnailFaceId != null)
+        'customThumbnailFaceId': customThumbnailFaceId,
+      'feedbackID': feedbackID,
+      'timestamp': timestamp.toIso8601String(),
+      'madeOnFaceMlVersion': madeOnFaceMlVersion,
+      'madeOnClusterMlVersion': madeOnClusterMlVersion,
+    };
+  }
+
+  @override
+  String toJsonString() => jsonEncode(toJson());
+
+  static RenameOrCustomThumbnailClusterFeedback fromJson(
+    Map<String, dynamic> json,
+  ) {
+    assert(
+      json['type'] ==
+          FeedbackType.renameOrCustomThumbnailClusterFeedback.toValueString(),
+    );
+    return RenameOrCustomThumbnailClusterFeedback(
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'],
+      customName: json['customName'],
+      customThumbnailFaceId: json['customThumbnailFaceId'],
+      feedbackID: json['feedbackID'],
+      timestamp: DateTime.parse(json['timestamp']),
+      madeOnFaceMlVersion: json['madeOnFaceMlVersion'],
+      madeOnClusterMlVersion: json['madeOnClusterMlVersion'],
+    );
+  }
+
+  static RenameOrCustomThumbnailClusterFeedback fromJsonString(
+    String jsonString,
+  ) {
+    return fromJson(jsonDecode(jsonString));
+  }
+}
+
+class RemovePhotosClusterFeedback extends ClusterFeedback {
+  final List<int> removedPhotosFileID;
+
+  RemovePhotosClusterFeedback({
+    required List<double> medoid,
+    required double medoidDistanceThreshold,
+    required this.removedPhotosFileID,
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  }) : super(
+          FeedbackType.removePhotosClusterFeedback,
+          medoid,
+          medoidDistanceThreshold,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  @override
+  Map<String, dynamic> toJson() {
+    return {
+      'type': type.toValueString(),
+      'medoid': medoid,
+      'medoidDistanceThreshold': medoidDistanceThreshold,
+      'removedPhotosFileID': removedPhotosFileID,
+      'feedbackID': feedbackID,
+      'timestamp': timestamp.toIso8601String(),
+      'madeOnFaceMlVersion': madeOnFaceMlVersion,
+      'madeOnClusterMlVersion': madeOnClusterMlVersion,
+    };
+  }
+
+  @override
+  String toJsonString() => jsonEncode(toJson());
+
+  static RemovePhotosClusterFeedback fromJson(Map<String, dynamic> json) {
+    assert(
+      json['type'] == FeedbackType.removePhotosClusterFeedback.toValueString(),
+    );
+    return RemovePhotosClusterFeedback(
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'],
+      removedPhotosFileID: (json['removedPhotosFileID'] as List?)
+              ?.map((item) => item as int)
+              .toList() ??
+          [],
+      feedbackID: json['feedbackID'],
+      timestamp: DateTime.parse(json['timestamp']),
+      madeOnFaceMlVersion: json['madeOnFaceMlVersion'],
+      madeOnClusterMlVersion: json['madeOnClusterMlVersion'],
+    );
+  }
+
+  static RemovePhotosClusterFeedback fromJsonString(String jsonString) {
+    return fromJson(jsonDecode(jsonString));
+  }
+}
+
+class AddPhotosClusterFeedback extends ClusterFeedback {
+  final List<int> addedPhotoFileIDs;
+
+  AddPhotosClusterFeedback({
+    required List<double> medoid,
+    required double medoidDistanceThreshold,
+    required this.addedPhotoFileIDs,
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  }) : super(
+          FeedbackType.addPhotosClusterFeedback,
+          medoid,
+          medoidDistanceThreshold,
+          feedbackID: feedbackID,
+          timestamp: timestamp,
+          madeOnFaceMlVersion: madeOnFaceMlVersion,
+          madeOnClusterMlVersion: madeOnClusterMlVersion,
+        );
+
+  @override
+  Map<String, dynamic> toJson() {
+    return {
+      'type': type.toValueString(),
+      'medoid': medoid,
+      'medoidDistanceThreshold': medoidDistanceThreshold,
+      'addedPhotoFileIDs': addedPhotoFileIDs,
+      'feedbackID': feedbackID,
+      'timestamp': timestamp.toIso8601String(),
+      'madeOnFaceMlVersion': madeOnFaceMlVersion,
+      'madeOnClusterMlVersion': madeOnClusterMlVersion,
+    };
+  }
+
+  @override
+  String toJsonString() => jsonEncode(toJson());
+
+  static AddPhotosClusterFeedback fromJson(Map<String, dynamic> json) {
+    assert(
+      json['type'] == FeedbackType.addPhotosClusterFeedback.toValueString(),
+    );
+    return AddPhotosClusterFeedback(
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'],
+      addedPhotoFileIDs: (json['addedPhotoFileIDs'] as List?)
+              ?.map((item) => item as int)
+              .toList() ??
+          [],
+      feedbackID: json['feedbackID'],
+      timestamp: DateTime.parse(json['timestamp']),
+      madeOnFaceMlVersion: json['madeOnFaceMlVersion'],
+      madeOnClusterMlVersion: json['madeOnClusterMlVersion'],
+    );
+  }
+
+  static AddPhotosClusterFeedback fromJsonString(String jsonString) {
+    return fromJson(jsonDecode(jsonString));
+  }
+}

mobile/lib/services/face_ml/face_feedback.dart/face_feedback_service.dart

@@ -0,0 +1,416 @@
+import "package:logging/logging.dart";
+import "package:photos/db/ml_data_db.dart";
+import "package:photos/services/face_ml/face_detection/detection.dart";
+import "package:photos/services/face_ml/face_feedback.dart/cluster_feedback.dart";
+import "package:photos/services/face_ml/face_ml_result.dart";
+
+class FaceFeedbackService {
+  final _logger = Logger("FaceFeedbackService");
+
+  final _mlDatabase = MlDataDB.instance;
+
+  int executedFeedbackCount = 0;
+  final int _reclusterFeedbackThreshold = 10;
+
+  // singleton pattern
+  FaceFeedbackService._privateConstructor();
+  static final instance = FaceFeedbackService._privateConstructor();
+  factory FaceFeedbackService() => instance;
+
+  /// Returns the updated cluster after removing the given file from the given person's cluster.
+  ///
+  /// If the file is not in the cluster, returns null.
+  ///
+  /// The updated cluster is also updated in [MlDataDB].
+  Future<ClusterResult> removePhotosFromCluster(
+    List<int> fileIDs,
+    int personID,
+  ) async {
+    // TODO: check if photo was originally added to cluster by user. If so, we should remove that addition instead of changing the embedding, because there is no embedding...
+    _logger.info(
+      'removePhotoFromCluster called with fileIDs $fileIDs and personID $personID',
+    );
+
+    if (fileIDs.isEmpty) {
+      _logger.severe(
+        "No fileIDs given, unable to add photos to cluster!",
+      );
+      throw ArgumentError(
+        "No fileIDs given, unable to add photos to cluster!",
+      );
+    }
+
+    // Get the relevant cluster
+    final ClusterResult? cluster = await _mlDatabase.getClusterResult(personID);
+    if (cluster == null) {
+      _logger.severe(
+        "No cluster found for personID $personID, unable to remove photo from non-existent cluster!",
+      );
+      throw ArgumentError(
+        "No cluster found for personID $personID, unable to remove photo from non-existent cluster!",
+      );
+    }
+    // Get the relevant faceMlResults
+    final List<FaceMlResult> faceMlResults =
+        await _mlDatabase.getSelectedFaceMlResults(fileIDs);
+    if (faceMlResults.length != fileIDs.length) {
+      final List<int> foundFileIDs =
+          faceMlResults.map((faceMlResult) => faceMlResult.fileId).toList();
+      _logger.severe(
+        "Couldn't find all facemlresults for fileIDs $fileIDs, only found for $foundFileIDs. Unable to remove unindexed photos from cluster!",
+      );
+      throw ArgumentError(
+        "Couldn't find all facemlresults for fileIDs $fileIDs, only found for $foundFileIDs. Unable to remove unindexed photos from cluster!",
+      );
+    }
+
+    // Check if at least one of the files is in the cluster. If all files are already not in the cluster, return the cluster.
+    final List<int> fileIDsInCluster = fileIDs
+        .where((fileID) => cluster.uniqueFileIds.contains(fileID))
+        .toList();
+    if (fileIDsInCluster.isEmpty) {
+      _logger.warning(
+        "All fileIDs are already not in the cluster, unable to remove photos from cluster!",
+      );
+      return cluster;
+    }
+    final List<FaceMlResult> faceMlResultsInCluster = faceMlResults
+        .where((faceMlResult) => fileIDsInCluster.contains(faceMlResult.fileId))
+        .toList();
+    assert(faceMlResultsInCluster.length == fileIDsInCluster.length);
+
+    for (var i = 0; i < fileIDsInCluster.length; i++) {
+      // Find the faces/embeddings associated with both the fileID and personID
+      final List<String> faceIDs = faceMlResultsInCluster[i].allFaceIds;
+      final List<String> faceIDsInCluster = cluster.faceIDs;
+      final List<String> relevantFaceIDs =
+          faceIDsInCluster.where((faceID) => faceIDs.contains(faceID)).toList();
+      if (relevantFaceIDs.isEmpty) {
+        _logger.severe(
+          "No faces found in both cluster and file, unable to remove photo from cluster!",
+        );
+        throw ArgumentError(
+          "No faces found in both cluster and file, unable to remove photo from cluster!",
+        );
+      }
+
+      // Set the embeddings to [10, 10,..., 10] and save the updated faceMlResult
+      faceMlResultsInCluster[i].setEmbeddingsToTen(relevantFaceIDs);
+      await _mlDatabase.updateFaceMlResult(faceMlResultsInCluster[i]);
+
+      // Make sure there is a manual override for [10, 10,..., 10] embeddings (not actually here, but in building the clusters, see _checkIfClusterIsDeleted function)
+
+      // Manually remove the fileID from the cluster
+      cluster.removeFileId(fileIDsInCluster[i]);
+    }
+
+    // TODO: see below
+    // Re-cluster and check if this leads to more deletions. If so, save them and ask the user if they want to delete them too.
+    executedFeedbackCount++;
+    if (executedFeedbackCount % _reclusterFeedbackThreshold == 0) {
+      // await recluster();
+    }
+
+    // Update the cluster in the database
+    await _mlDatabase.updateClusterResult(cluster);
+
+    // TODO: see below
+    // Safe the given feedback to the database
+    final removePhotoFeedback = RemovePhotosClusterFeedback(
+      medoid: cluster.medoid,
+      medoidDistanceThreshold: cluster.medoidDistanceThreshold,
+      removedPhotosFileID: fileIDsInCluster,
+    );
+    await _mlDatabase.createClusterFeedback(
+      removePhotoFeedback,
+      skipIfSimilarFeedbackExists: false,
+    );
+
+    // Return the updated cluster
+    return cluster;
+  }
+
+  Future<ClusterResult> addPhotosToCluster(List<int> fileIDs, personID) async {
+    _logger.info(
+      'addPhotosToCluster called with fileIDs $fileIDs and personID $personID',
+    );
+
+    if (fileIDs.isEmpty) {
+      _logger.severe(
+        "No fileIDs given, unable to add photos to cluster!",
+      );
+      throw ArgumentError(
+        "No fileIDs given, unable to add photos to cluster!",
+      );
+    }
+
+    // Get the relevant cluster
+    final ClusterResult? cluster = await _mlDatabase.getClusterResult(personID);
+    if (cluster == null) {
+      _logger.severe(
+        "No cluster found for personID $personID, unable to add photos to non-existent cluster!",
+      );
+      throw ArgumentError(
+        "No cluster found for personID $personID, unable to add photos to non-existent cluster!",
+      );
+    }
+
+    // Check if at least one of the files is not in the cluster. If all files are already in the cluster, return the cluster.
+    final List<int> fileIDsNotInCluster = fileIDs
+        .where((fileID) => !cluster.uniqueFileIds.contains(fileID))
+        .toList();
+    if (fileIDsNotInCluster.isEmpty) {
+      _logger.warning(
+        "All fileIDs are already in the cluster, unable to add new photos to cluster!",
+      );
+      return cluster;
+    }
+    final List<String> faceIDsNotInCluster = fileIDsNotInCluster
+        .map((fileID) => FaceDetectionRelative.toFaceIDEmpty(fileID: fileID))
+        .toList();
+
+    // Add the new files to the cluster
+    cluster.addFileIDsAndFaceIDs(fileIDsNotInCluster, faceIDsNotInCluster);
+
+    // Update the cluster in the database
+    await _mlDatabase.updateClusterResult(cluster);
+
+    // Build the addPhotoFeedback
+    final AddPhotosClusterFeedback addPhotosFeedback = AddPhotosClusterFeedback(
+      medoid: cluster.medoid,
+      medoidDistanceThreshold: cluster.medoidDistanceThreshold,
+      addedPhotoFileIDs: fileIDsNotInCluster,
+    );
+
+    // TODO: check for exact match and update feedback if necessary
+
+    // Save the addPhotoFeedback to the database
+    await _mlDatabase.createClusterFeedback(
+      addPhotosFeedback,
+      skipIfSimilarFeedbackExists: false,
+    );
+
+    // Return the updated cluster
+    return cluster;
+  }
+
+  /// Deletes the given cluster completely.
+  Future<void> deleteCluster(int personID) async {
+    _logger.info(
+      'deleteCluster called with personID $personID',
+    );
+
+    // Get the relevant cluster
+    final cluster = await _mlDatabase.getClusterResult(personID);
+    if (cluster == null) {
+      _logger.severe(
+        "No cluster found for personID $personID, unable to delete non-existent cluster!",
+      );
+      throw ArgumentError(
+        "No cluster found for personID $personID, unable to delete non-existent cluster!",
+      );
+    }
+
+    // Delete the cluster from the database
+    await _mlDatabase.deleteClusterResult(cluster.personId);
+
+    // TODO: look into the right threshold distance.
+    // Build the deleteClusterFeedback
+    final DeleteClusterFeedback deleteClusterFeedback = DeleteClusterFeedback(
+      medoid: cluster.medoid,
+      medoidDistanceThreshold: cluster.medoidDistanceThreshold,
+    );
+
+    // TODO: maybe I should merge the two feedbacks if they are similar enough? Or alternatively, I keep them both?
+    // Check if feedback doesn't already exist
+    if (await _mlDatabase
+        .doesSimilarClusterFeedbackExist(deleteClusterFeedback)) {
+      _logger.warning(
+        "Feedback already exists for deleting cluster $personID, unable to delete cluster!",
+      );
+      return;
+    }
+
+    // Save the deleteClusterFeedback to the database
+    await _mlDatabase.createClusterFeedback(deleteClusterFeedback);
+  }
+
+  /// Renames the given cluster and/or sets the thumbnail of the given cluster.
+  ///
+  /// Requires either a [customName] or a [customFaceID]. If both are given, both are used. If neither are given, an error is thrown.
+  Future<ClusterResult> renameOrSetThumbnailCluster(
+    int personID, {
+    String? customName,
+    String? customFaceID,
+  }) async {
+    _logger.info(
+      'renameOrSetThumbnailCluster called with personID $personID, customName $customName, and customFaceID $customFaceID',
+    );
+
+    if (customFaceID != null &&
+        FaceDetectionRelative.isFaceIDEmpty(customFaceID)) {
+      _logger.severe(
+        "customFaceID $customFaceID is belongs to empty detection, unable to set as thumbnail of cluster!",
+      );
+      customFaceID = null;
+    }
+    if (customName == null && customFaceID == null) {
+      _logger.severe(
+        "No name or faceID given, unable to rename or set thumbnail of cluster!",
+      );
+      throw ArgumentError(
+        "No name or faceID given, unable to rename or set thumbnail of cluster!",
+      );
+    }
+
+    // Get the relevant cluster
+    final cluster = await _mlDatabase.getClusterResult(personID);
+    if (cluster == null) {
+      _logger.severe(
+        "No cluster found for personID $personID, unable to delete non-existent cluster!",
+      );
+      throw ArgumentError(
+        "No cluster found for personID $personID, unable to delete non-existent cluster!",
+      );
+    }
+
+    // Update the cluster
+    if (customName != null) cluster.setUserDefinedName = customName;
+    if (customFaceID != null) cluster.setThumbnailFaceId = customFaceID;
+
+    // Update the cluster in the database
+    await _mlDatabase.updateClusterResult(cluster);
+
+    // Build the RenameOrCustomThumbnailClusterFeedback
+    final RenameOrCustomThumbnailClusterFeedback renameClusterFeedback =
+        RenameOrCustomThumbnailClusterFeedback(
+      medoid: cluster.medoid,
+      medoidDistanceThreshold: cluster.medoidDistanceThreshold,
+      customName: customName,
+      customThumbnailFaceId: customFaceID,
+    );
+
+    // TODO: maybe I should merge the two feedbacks if they are similar enough?
+    // Check if feedback doesn't already exist
+    final matchingFeedbacks =
+        await _mlDatabase.getAllMatchingClusterFeedback(renameClusterFeedback);
+    for (final matchingFeedback in matchingFeedbacks) {
+      // Update the current feedback wherever possible
+      renameClusterFeedback.customName ??= matchingFeedback.customName;
+      renameClusterFeedback.customThumbnailFaceId ??=
+          matchingFeedback.customThumbnailFaceId;
+
+      // Delete the old feedback (since we want the user to be able to overwrite their earlier feedback)
+      await _mlDatabase.deleteClusterFeedback(matchingFeedback);
+    }
+
+    // Save the RenameOrCustomThumbnailClusterFeedback to the database
+    await _mlDatabase.createClusterFeedback(renameClusterFeedback);
+
+    // Return the updated cluster
+    return cluster;
+  }
+
+  /// Merges the given clusters. The largest cluster is kept and the other clusters are deleted.
+  ///
+  /// Requires either a [clusters] or [personIDs]. If both are given, the [clusters] are used.
+  Future<ClusterResult> mergeClusters(List<int> personIDs) async {
+    _logger.info(
+      'mergeClusters called with personIDs $personIDs',
+    );
+
+    // Get the relevant clusters
+    final List<ClusterResult> clusters =
+        await _mlDatabase.getSelectedClusterResults(personIDs);
+    if (clusters.length <= 1) {
+      _logger.severe(
+        "${clusters.length} clusters found for personIDs $personIDs, unable to merge non-existent clusters!",
+      );
+      throw ArgumentError(
+        "${clusters.length} clusters found for personIDs $personIDs, unable to merge non-existent clusters!",
+      );
+    }
+
+    // Find the largest cluster
+    clusters.sort((a, b) => b.clusterSize.compareTo(a.clusterSize));
+    final ClusterResult largestCluster = clusters.first;
+
+    // Now iterate through the clusters to be merged and deleted
+    for (var i = 1; i < clusters.length; i++) {
+      final ClusterResult clusterToBeMerged = clusters[i];
+
+      // Add the files and faces of the cluster to be merged to the largest cluster
+      largestCluster.addFileIDsAndFaceIDs(
+        clusterToBeMerged.fileIDsIncludingPotentialDuplicates,
+        clusterToBeMerged.faceIDs,
+      );
+
+      // TODO: maybe I should wrap the logic below in a separate function, since it's also used in renameOrSetThumbnailCluster
+      // Merge any names and thumbnails if the largest cluster doesn't have them
+      bool shouldCreateNamingFeedback = false;
+      String? nameToBeMerged;
+      String? thumbnailToBeMerged;
+      if (!largestCluster.hasUserDefinedName &&
+          clusterToBeMerged.hasUserDefinedName) {
+        largestCluster.setUserDefinedName = clusterToBeMerged.userDefinedName!;
+        nameToBeMerged = clusterToBeMerged.userDefinedName!;
+        shouldCreateNamingFeedback = true;
+      }
+      if (!largestCluster.thumbnailFaceIdIsUserDefined &&
+          clusterToBeMerged.thumbnailFaceIdIsUserDefined) {
+        largestCluster.setThumbnailFaceId = clusterToBeMerged.thumbnailFaceId;
+        thumbnailToBeMerged = clusterToBeMerged.thumbnailFaceId;
+        shouldCreateNamingFeedback = true;
+      }
+      if (shouldCreateNamingFeedback) {
+        final RenameOrCustomThumbnailClusterFeedback renameClusterFeedback =
+            RenameOrCustomThumbnailClusterFeedback(
+          medoid: largestCluster.medoid,
+          medoidDistanceThreshold: largestCluster.medoidDistanceThreshold,
+          customName: nameToBeMerged,
+          customThumbnailFaceId: thumbnailToBeMerged,
+        );
+        // Check if feedback doesn't already exist
+        final matchingFeedbacks = await _mlDatabase
+            .getAllMatchingClusterFeedback(renameClusterFeedback);
+        for (final matchingFeedback in matchingFeedbacks) {
+          // Update the current feedback wherever possible
+          renameClusterFeedback.customName ??= matchingFeedback.customName;
+          renameClusterFeedback.customThumbnailFaceId ??=
+              matchingFeedback.customThumbnailFaceId;
+
+          // Delete the old feedback (since we want the user to be able to overwrite their earlier feedback)
+          await _mlDatabase.deleteClusterFeedback(matchingFeedback);
+        }
+
+        // Save the RenameOrCustomThumbnailClusterFeedback to the database
+        await _mlDatabase.createClusterFeedback(renameClusterFeedback);
+      }
+
+      // Build the mergeClusterFeedback
+      final MergeClusterFeedback mergeClusterFeedback = MergeClusterFeedback(
+        medoid: clusterToBeMerged.medoid,
+        medoidDistanceThreshold: clusterToBeMerged.medoidDistanceThreshold,
+        medoidToMoveTo: largestCluster.medoid,
+      );
+
+      // Save the mergeClusterFeedback to the database and delete any old matching feedbacks
+      final matchingFeedbacks =
+          await _mlDatabase.getAllMatchingClusterFeedback(mergeClusterFeedback);
+      for (final matchingFeedback in matchingFeedbacks) {
+        await _mlDatabase.deleteClusterFeedback(matchingFeedback);
+      }
+      await _mlDatabase.createClusterFeedback(mergeClusterFeedback);
+
+      // Delete the cluster from the database
+      await _mlDatabase.deleteClusterResult(clusterToBeMerged.personId);
+    }
+
+    // TODO: should I update the medoid of this new cluster? My intuition says no, but I'm not sure.
+    // Update the largest cluster in the database
+    await _mlDatabase.updateClusterResult(largestCluster);
+
+    // Return the merged cluster
+    return largestCluster;
+  }
+}

mobile/lib/services/face_ml/face_feedback.dart/feedback.dart

@@ -0,0 +1,34 @@
+import "package:photos/models/ml/ml_versions.dart";
+import "package:photos/services/face_ml/face_feedback.dart/feedback_types.dart";
+import "package:uuid/uuid.dart";
+
+abstract class Feedback {
+  final FeedbackType type;
+  final String feedbackID;
+  final DateTime timestamp;
+  final int madeOnFaceMlVersion;
+  final int madeOnClusterMlVersion;
+
+  get typeString => type.toValueString();
+
+  get timestampString => timestamp.toIso8601String();
+
+  Feedback(
+    this.type, {
+    String? feedbackID,
+    DateTime? timestamp,
+    int? madeOnFaceMlVersion,
+    int? madeOnClusterMlVersion,
+  })  : feedbackID = feedbackID ?? const Uuid().v4(),
+        timestamp = timestamp ?? DateTime.now(),
+        madeOnFaceMlVersion = madeOnFaceMlVersion ?? faceMlVersion,
+        madeOnClusterMlVersion = madeOnClusterMlVersion ?? clusterMlVersion;
+
+  Map<String, dynamic> toJson();
+
+  String toJsonString();
+
+  // Feedback fromJson(Map<String, dynamic> json);
+
+  // Feedback fromJsonString(String jsonString);
+}

mobile/lib/services/face_ml/face_feedback.dart/feedback_types.dart

@@ -0,0 +1,26 @@
+enum FeedbackType {
+  removePhotosClusterFeedback,
+  addPhotosClusterFeedback,
+  deleteClusterFeedback,
+  mergeClusterFeedback,
+  renameOrCustomThumbnailClusterFeedback; // I have merged renameClusterFeedback and customThumbnailClusterFeedback, since I suspect they will be used together often
+
+  factory FeedbackType.fromValueString(String value) {
+    switch (value) {
+      case 'deleteClusterFeedback':
+        return FeedbackType.deleteClusterFeedback;
+      case 'mergeClusterFeedback':
+        return FeedbackType.mergeClusterFeedback;
+      case 'renameOrCustomThumbnailClusterFeedback':
+        return FeedbackType.renameOrCustomThumbnailClusterFeedback;
+      case 'removePhotoClusterFeedback':
+        return FeedbackType.removePhotosClusterFeedback;
+      case 'addPhotoClusterFeedback':
+        return FeedbackType.addPhotosClusterFeedback;
+      default:
+        throw Exception('Invalid FeedbackType: $value');
+    }
+  }
+
+  String toValueString() => name;
+}

mobile/lib/services/face_ml/face_ml_exceptions.dart

@@ -0,0 +1,30 @@
+
+class GeneralFaceMlException implements Exception {
+  final String message;
+
+  GeneralFaceMlException(this.message);
+
+  @override
+  String toString() => 'GeneralFaceMlException: $message';
+}
+
+class CouldNotRetrieveAnyFileData implements Exception {}
+
+class CouldNotInitializeFaceDetector implements Exception {}
+
+class CouldNotRunFaceDetector implements Exception {}
+
+class CouldNotWarpAffine implements Exception {}
+
+class CouldNotInitializeFaceEmbeddor implements Exception {}
+
+class InputProblemFaceEmbeddor implements Exception {
+  final String message;
+
+  InputProblemFaceEmbeddor(this.message);
+
+  @override
+  String toString() => 'InputProblemFaceEmbeddor: $message';
+}
+
+class CouldNotRunFaceEmbeddor implements Exception {}

mobile/lib/services/face_ml/face_ml_methods.dart

@@ -0,0 +1,90 @@
+import "package:photos/services/face_ml/face_ml_version.dart";
+
+/// Represents a face detection method with a specific version.
+class FaceDetectionMethod extends VersionedMethod {
+  /// Creates a [FaceDetectionMethod] instance with a specific `method` and `version` (default `1`)
+  FaceDetectionMethod(String method, {int version = 1})
+      : super(method, version);
+
+  /// Creates a [FaceDetectionMethod] instance with 'Empty method' as the method, and a specific `version` (default `1`)
+  const FaceDetectionMethod.empty() : super.empty();
+
+  /// Creates a [FaceDetectionMethod] instance with 'BlazeFace' as the method, and a specific `version` (default `1`)
+  FaceDetectionMethod.blazeFace({int version = 1})
+      : super('BlazeFace', version);
+
+  static FaceDetectionMethod fromMlVersion(int version) {
+    switch (version) {
+      case 1:
+        return FaceDetectionMethod.blazeFace(version: version);
+      default:
+        return const FaceDetectionMethod.empty();
+    }
+  }
+
+  static FaceDetectionMethod fromJson(Map<String, dynamic> json) {
+    return FaceDetectionMethod(
+      json['method'],
+      version: json['version'],
+    );
+  }
+}
+
+/// Represents a face alignment method with a specific version.
+class FaceAlignmentMethod extends VersionedMethod {
+  /// Creates a [FaceAlignmentMethod] instance with a specific `method` and `version` (default `1`)
+  FaceAlignmentMethod(String method, {int version = 1})
+      : super(method, version);
+
+  /// Creates a [FaceAlignmentMethod] instance with 'Empty method' as the method, and a specific `version` (default `1`)
+  const FaceAlignmentMethod.empty() : super.empty();
+
+  /// Creates a [FaceAlignmentMethod] instance with 'ArcFace' as the method, and a specific `version` (default `1`)
+  FaceAlignmentMethod.arcFace({int version = 1}) : super('ArcFace', version);
+
+  static FaceAlignmentMethod fromMlVersion(int version) {
+    switch (version) {
+      case 1:
+        return FaceAlignmentMethod.arcFace(version: version);
+      default:
+        return const FaceAlignmentMethod.empty();
+    }
+  }
+
+  static FaceAlignmentMethod fromJson(Map<String, dynamic> json) {
+    return FaceAlignmentMethod(
+      json['method'],
+      version: json['version'],
+    );
+  }
+}
+
+/// Represents a face embedding method with a specific version.
+class FaceEmbeddingMethod extends VersionedMethod {
+  /// Creates a [FaceEmbeddingMethod] instance with a specific `method` and `version` (default `1`)
+  FaceEmbeddingMethod(String method, {int version = 1})
+      : super(method, version);
+
+  /// Creates a [FaceEmbeddingMethod] instance with 'Empty method' as the method, and a specific `version` (default `1`)
+  const FaceEmbeddingMethod.empty() : super.empty();
+
+  /// Creates a [FaceEmbeddingMethod] instance with 'MobileFaceNet' as the method, and a specific `version` (default `1`)
+  FaceEmbeddingMethod.mobileFaceNet({int version = 1})
+      : super('MobileFaceNet', version);
+
+  static FaceEmbeddingMethod fromMlVersion(int version) {
+    switch (version) {
+      case 1:
+        return FaceEmbeddingMethod.mobileFaceNet(version: version);
+      default:
+        return const FaceEmbeddingMethod.empty();
+    }
+  }
+
+  static FaceEmbeddingMethod fromJson(Map<String, dynamic> json) {
+    return FaceEmbeddingMethod(
+      json['method'],
+      version: json['version'],
+    );
+  }
+}

mobile/lib/services/face_ml/face_ml_result.dart

@@ -0,0 +1,753 @@
+import "dart:convert" show jsonEncode, jsonDecode;
+
+import "package:flutter/material.dart" show Size, debugPrint, immutable;
+import "package:logging/logging.dart";
+import "package:photos/db/ml_data_db.dart";
+import "package:photos/models/file/file.dart";
+import 'package:photos/models/ml/ml_typedefs.dart';
+import "package:photos/models/ml/ml_versions.dart";
+import "package:photos/services/face_ml/blur_detection/blur_constants.dart";
+import "package:photos/services/face_ml/face_alignment/alignment_result.dart";
+import "package:photos/services/face_ml/face_clustering/cosine_distance.dart";
+import "package:photos/services/face_ml/face_detection/detection.dart";
+import "package:photos/services/face_ml/face_feedback.dart/cluster_feedback.dart";
+import "package:photos/services/face_ml/face_ml_methods.dart";
+
+final _logger = Logger('ClusterResult_FaceMlResult');
+
+// TODO: should I add [faceMlVersion] and [clusterMlVersion] to the [ClusterResult] class?
+class ClusterResult {
+  final int personId;
+  String? userDefinedName;
+  bool get hasUserDefinedName => userDefinedName != null;
+
+  String _thumbnailFaceId;
+  bool thumbnailFaceIdIsUserDefined;
+
+  final List<int> _fileIds;
+  final List<String> _faceIds;
+
+  final Embedding medoid;
+  double medoidDistanceThreshold;
+
+  List<int> get uniqueFileIds => _fileIds.toSet().toList();
+  List<int> get fileIDsIncludingPotentialDuplicates => _fileIds;
+
+  List<String> get faceIDs => _faceIds;
+
+  String get thumbnailFaceId => _thumbnailFaceId;
+
+  int get thumbnailFileId => _getFileIdFromFaceId(_thumbnailFaceId);
+
+  /// Sets the thumbnail faceId to the given faceId.
+  /// Throws an exception if the faceId is not in the list of faceIds.
+  set setThumbnailFaceId(String faceId) {
+    if (!_faceIds.contains(faceId)) {
+      throw Exception(
+        "The faceId $faceId is not in the list of faceIds: $faceId",
+      );
+    }
+    _thumbnailFaceId = faceId;
+    thumbnailFaceIdIsUserDefined = true;
+  }
+
+  /// Sets the [userDefinedName] to the given [customName]
+  set setUserDefinedName(String customName) {
+    userDefinedName = customName;
+  }
+
+  int get clusterSize => _fileIds.toSet().length;
+
+  ClusterResult({
+    required this.personId,
+    required String thumbnailFaceId,
+    required List<int> fileIds,
+    required List<String> faceIds,
+    required this.medoid,
+    required this.medoidDistanceThreshold,
+    this.userDefinedName,
+    this.thumbnailFaceIdIsUserDefined = false,
+  })  : _thumbnailFaceId = thumbnailFaceId,
+        _faceIds = faceIds,
+        _fileIds = fileIds;
+
+  void addFileIDsAndFaceIDs(List<int> fileIDs, List<String> faceIDs) {
+    assert(fileIDs.length == faceIDs.length);
+    _fileIds.addAll(fileIDs);
+    _faceIds.addAll(faceIDs);
+  }
+
+  // TODO: Consider if we should recalculated the medoid and threshold when deleting or adding a file from the cluster
+  int removeFileId(int fileId) {
+    assert(_fileIds.length == _faceIds.length);
+    if (!_fileIds.contains(fileId)) {
+      throw Exception(
+        "The fileId $fileId is not in the list of fileIds: $fileId, so it's not in the cluster and cannot be removed.",
+      );
+    }
+
+    int removedCount = 0;
+    for (var i = 0; i < _fileIds.length; i++) {
+      if (_fileIds[i] == fileId) {
+        assert(_getFileIdFromFaceId(_faceIds[i]) == fileId);
+        _fileIds.removeAt(i);
+        _faceIds.removeAt(i);
+        debugPrint(
+          "Removed fileId $fileId from cluster $personId at index ${i + removedCount}}",
+        );
+        i--; // Adjust index due to removal
+        removedCount++;
+      }
+    }
+
+    _ensureClusterSizeIsAboveMinimum();
+
+    return removedCount;
+  }
+
+  int addFileID(int fileID) {
+    assert(_fileIds.length == _faceIds.length);
+    if (_fileIds.contains(fileID)) {
+      return 0;
+    }
+
+    _fileIds.add(fileID);
+    _faceIds.add(FaceDetectionRelative.toFaceIDEmpty(fileID: fileID));
+
+    return 1;
+  }
+
+  void ensureThumbnailFaceIdIsInCluster() {
+    if (!_faceIds.contains(_thumbnailFaceId)) {
+      _thumbnailFaceId = _faceIds[0];
+    }
+  }
+
+  void _ensureClusterSizeIsAboveMinimum() {
+    if (clusterSize < minimumClusterSize) {
+      throw Exception(
+        "Cluster size is below minimum cluster size of $minimumClusterSize",
+      );
+    }
+  }
+
+  Map<String, dynamic> _toJson() => {
+        'personId': personId,
+        'thumbnailFaceId': _thumbnailFaceId,
+        'fileIds': _fileIds,
+        'faceIds': _faceIds,
+        'medoid': medoid,
+        'medoidDistanceThreshold': medoidDistanceThreshold,
+        if (userDefinedName != null) 'userDefinedName': userDefinedName,
+        'thumbnailFaceIdIsUserDefined': thumbnailFaceIdIsUserDefined,
+      };
+
+  String toJsonString() => jsonEncode(_toJson());
+
+  static ClusterResult _fromJson(Map<String, dynamic> json) {
+    return ClusterResult(
+      personId: json['personId'] ?? -1,
+      thumbnailFaceId: json['thumbnailFaceId'] ?? '',
+      fileIds:
+          (json['fileIds'] as List?)?.map((item) => item as int).toList() ?? [],
+      faceIds:
+          (json['faceIds'] as List?)?.map((item) => item as String).toList() ??
+              [],
+      medoid:
+          (json['medoid'] as List?)?.map((item) => item as double).toList() ??
+              [],
+      medoidDistanceThreshold: json['medoidDistanceThreshold'] ?? 0,
+      userDefinedName: json['userDefinedName'],
+      thumbnailFaceIdIsUserDefined:
+          json['thumbnailFaceIdIsUserDefined'] as bool,
+    );
+  }
+
+  static ClusterResult fromJsonString(String jsonString) {
+    return _fromJson(jsonDecode(jsonString));
+  }
+}
+
+class ClusterResultBuilder {
+  int personId = -1;
+  String? userDefinedName;
+  String thumbnailFaceId = '';
+  bool thumbnailFaceIdIsUserDefined = false;
+
+  List<int> fileIds = <int>[];
+  List<String> faceIds = <String>[];
+
+  List<Embedding> embeddings = <Embedding>[];
+  Embedding medoid = <double>[];
+  double medoidDistanceThreshold = 0;
+  bool medoidAndThresholdCalculated = false;
+  final int k = 5;
+
+  ClusterResultBuilder.createFromIndices({
+    required List<int> clusterIndices,
+    required List<int> labels,
+    required List<Embedding> allEmbeddings,
+    required List<int> allFileIds,
+    required List<String> allFaceIds,
+  }) {
+    final clusteredFileIds =
+        clusterIndices.map((fileIndex) => allFileIds[fileIndex]).toList();
+    final clusteredFaceIds =
+        clusterIndices.map((fileIndex) => allFaceIds[fileIndex]).toList();
+    final clusteredEmbeddings =
+        clusterIndices.map((fileIndex) => allEmbeddings[fileIndex]).toList();
+    personId = labels[clusterIndices[0]];
+    fileIds = clusteredFileIds;
+    faceIds = clusteredFaceIds;
+    thumbnailFaceId = faceIds[0];
+    embeddings = clusteredEmbeddings;
+  }
+
+  void calculateAndSetMedoidAndThreshold() {
+    if (embeddings.isEmpty) {
+      throw Exception("Cannot calculate medoid and threshold for empty list");
+    }
+
+    // Calculate the medoid and threshold
+    final (tempMedoid, distanceThreshold) =
+        _calculateMedoidAndDistanceTreshold(embeddings);
+
+    // Update the medoid
+    medoid = List.from(tempMedoid);
+
+    // Update the medoidDistanceThreshold as the distance of the medoid to its k-th nearest neighbor
+    medoidDistanceThreshold = distanceThreshold;
+
+    medoidAndThresholdCalculated = true;
+  }
+
+  (List<double>, double) _calculateMedoidAndDistanceTreshold(
+    List<List<double>> embeddings,
+  ) {
+    double minDistance = double.infinity;
+    List<double>? medoid;
+
+    // Calculate the distance between all pairs
+    for (int i = 0; i < embeddings.length; ++i) {
+      double totalDistance = 0;
+      for (int j = 0; j < embeddings.length; ++j) {
+        if (i != j) {
+          totalDistance += cosineDistance(embeddings[i], embeddings[j]);
+
+          // Break early if we already exceed minDistance
+          if (totalDistance > minDistance) {
+            break;
+          }
+        }
+      }
+
+      // Find the minimum total distance
+      if (totalDistance < minDistance) {
+        minDistance = totalDistance;
+        medoid = embeddings[i];
+      }
+    }
+
+    // Now, calculate k-th nearest neighbor for the medoid
+    final List<double> distancesToMedoid = [];
+    for (List<double> embedding in embeddings) {
+      if (embedding != medoid) {
+        distancesToMedoid.add(cosineDistance(medoid!, embedding));
+      }
+    }
+    distancesToMedoid.sort();
+    // TODO: empirically find the best k. Probably it should be dynamic in some way, so for instance larger for larger clusters and smaller for smaller clusters, especially since there are a lot of really small clusters and a few really large ones.
+    final double kthDistance = distancesToMedoid[
+        distancesToMedoid.length >= k ? k - 1 : distancesToMedoid.length - 1];
+
+    return (medoid!, kthDistance);
+  }
+
+  Future<bool> _checkIfClusterIsDeleted() async {
+    assert(medoidAndThresholdCalculated);
+
+    // Check if the medoid is the default medoid for deleted faces
+    if (cosineDistance(medoid, List.filled(medoid.length, 10.0)) < 0.001) {
+      return true;
+    }
+
+    final tempFeedback = DeleteClusterFeedback(
+      medoid: medoid,
+      medoidDistanceThreshold: medoidDistanceThreshold,
+    );
+    return await MlDataDB.instance
+        .doesSimilarClusterFeedbackExist(tempFeedback);
+  }
+
+  Future<void> _checkAndAddPhotos() async {
+    assert(medoidAndThresholdCalculated);
+
+    final tempFeedback = AddPhotosClusterFeedback(
+      medoid: medoid,
+      medoidDistanceThreshold: medoidDistanceThreshold,
+      addedPhotoFileIDs: [],
+    );
+    final allAddPhotosFeedbacks =
+        await MlDataDB.instance.getAllMatchingClusterFeedback(tempFeedback);
+
+    for (final addPhotosFeedback in allAddPhotosFeedbacks) {
+      final fileIDsToAdd = addPhotosFeedback.addedPhotoFileIDs;
+      final faceIDsToAdd = fileIDsToAdd
+          .map((fileID) => FaceDetectionRelative.toFaceIDEmpty(fileID: fileID))
+          .toList();
+      addFileIDsAndFaceIDs(fileIDsToAdd, faceIDsToAdd);
+    }
+  }
+
+  Future<void> _checkAndAddCustomName() async {
+    assert(medoidAndThresholdCalculated);
+
+    final tempFeedback = RenameOrCustomThumbnailClusterFeedback(
+      medoid: medoid,
+      medoidDistanceThreshold: medoidDistanceThreshold,
+      customName: 'test',
+    );
+    final allRenameFeedbacks =
+        await MlDataDB.instance.getAllMatchingClusterFeedback(tempFeedback);
+
+    for (final nameFeedback in allRenameFeedbacks) {
+      userDefinedName ??= nameFeedback.customName;
+      if (!thumbnailFaceIdIsUserDefined) {
+        thumbnailFaceId = nameFeedback.customThumbnailFaceId ?? thumbnailFaceId;
+        thumbnailFaceIdIsUserDefined =
+            nameFeedback.customThumbnailFaceId != null;
+      }
+    }
+    return;
+  }
+
+  void changeThumbnailFaceId(String faceId) {
+    if (!faceIds.contains(faceId)) {
+      throw Exception(
+        "The faceId $faceId is not in the list of faceIds: $faceIds",
+      );
+    }
+    thumbnailFaceId = faceId;
+  }
+
+  void addFileIDsAndFaceIDs(List<int> addedFileIDs, List<String> addedFaceIDs) {
+    assert(addedFileIDs.length == addedFaceIDs.length);
+    fileIds.addAll(addedFileIDs);
+    faceIds.addAll(addedFaceIDs);
+  }
+
+  static Future<List<ClusterResult>> buildClusters(
+    List<ClusterResultBuilder> clusterBuilders,
+  ) async {
+    final List<int> deletedClusterIndices = [];
+    for (var i = 0; i < clusterBuilders.length; i++) {
+      final clusterBuilder = clusterBuilders[i];
+      clusterBuilder.calculateAndSetMedoidAndThreshold();
+
+      // Check if the cluster has been deleted
+      if (await clusterBuilder._checkIfClusterIsDeleted()) {
+        deletedClusterIndices.add(i);
+      }
+
+      await clusterBuilder._checkAndAddPhotos();
+    }
+
+    // Check if a cluster should be merged with another cluster
+    for (var i = 0; i < clusterBuilders.length; i++) {
+      // Don't check for clusters that have been deleted
+      if (deletedClusterIndices.contains(i)) {
+        continue;
+      }
+      final clusterBuilder = clusterBuilders[i];
+      final List<MergeClusterFeedback> allMatchingMergeFeedback =
+          await MlDataDB.instance.getAllMatchingClusterFeedback(
+        MergeClusterFeedback(
+          medoid: clusterBuilder.medoid,
+          medoidDistanceThreshold: clusterBuilder.medoidDistanceThreshold,
+          medoidToMoveTo: clusterBuilder.medoid,
+        ),
+      );
+      if (allMatchingMergeFeedback.isEmpty) {
+        continue;
+      }
+      // Merge the cluster with the first merge feedback
+      final mainFeedback = allMatchingMergeFeedback.first;
+      if (allMatchingMergeFeedback.length > 1) {
+        // This is the BUG!!!!
+        _logger.warning(
+          "There are ${allMatchingMergeFeedback.length} merge feedbacks for cluster ${clusterBuilder.personId}. Using the first one.",
+        );
+      }
+      for (var j = 0; j < clusterBuilders.length; j++) {
+        if (i == j) continue;
+        final clusterBuilderToMergeTo = clusterBuilders[j];
+        final distance = cosineDistance(
+          // BUG: it hasn't calculated the medoid for every clusterBuilder yet!!!
+          mainFeedback.medoidToMoveTo,
+          clusterBuilderToMergeTo.medoid,
+        );
+        if (distance < mainFeedback.medoidDistanceThreshold ||
+            distance < clusterBuilderToMergeTo.medoidDistanceThreshold) {
+          clusterBuilderToMergeTo.addFileIDsAndFaceIDs(
+            clusterBuilder.fileIds,
+            clusterBuilder.faceIds,
+          );
+          deletedClusterIndices.add(i);
+        }
+      }
+    }
+
+    final clusterResults = <ClusterResult>[];
+    for (var i = 0; i < clusterBuilders.length; i++) {
+      // Don't build the cluster if it has been deleted or merged
+      if (deletedClusterIndices.contains(i)) {
+        continue;
+      }
+      final clusterBuilder = clusterBuilders[i];
+      // Check if the cluster has a custom name or thumbnail
+      await clusterBuilder._checkAndAddCustomName();
+
+      // Build the clusterResult
+      clusterResults.add(
+        ClusterResult(
+          personId: clusterBuilder.personId,
+          thumbnailFaceId: clusterBuilder.thumbnailFaceId,
+          fileIds: clusterBuilder.fileIds,
+          faceIds: clusterBuilder.faceIds,
+          medoid: clusterBuilder.medoid,
+          medoidDistanceThreshold: clusterBuilder.medoidDistanceThreshold,
+          userDefinedName: clusterBuilder.userDefinedName,
+          thumbnailFaceIdIsUserDefined:
+              clusterBuilder.thumbnailFaceIdIsUserDefined,
+        ),
+      );
+    }
+
+    return clusterResults;
+  }
+
+  // TODO: This function should include the feedback from the user. Should also be nullable, since user might want to delete the cluster.
+  Future<ClusterResult?> _buildSingleCluster() async {
+    calculateAndSetMedoidAndThreshold();
+    if (await _checkIfClusterIsDeleted()) {
+      return null;
+    }
+    await _checkAndAddCustomName();
+    return ClusterResult(
+      personId: personId,
+      thumbnailFaceId: thumbnailFaceId,
+      fileIds: fileIds,
+      faceIds: faceIds,
+      medoid: medoid,
+      medoidDistanceThreshold: medoidDistanceThreshold,
+    );
+  }
+}
+
+@immutable
+class FaceMlResult {
+  final int fileId;
+
+  final List<FaceResult> faces;
+
+  final Size? faceDetectionImageSize;
+  final Size? faceAlignmentImageSize;
+
+  final int mlVersion;
+  final bool errorOccured;
+  final bool onlyThumbnailUsed;
+
+  bool get hasFaces => faces.isNotEmpty;
+  int get numberOfFaces => faces.length;
+
+  List<Embedding> get allFaceEmbeddings {
+    return faces.map((face) => face.embedding).toList();
+  }
+
+  List<String> get allFaceIds {
+    return faces.map((face) => face.faceId).toList();
+  }
+
+  List<int> get fileIdForEveryFace {
+    return List<int>.filled(faces.length, fileId);
+  }
+
+  FaceDetectionMethod get faceDetectionMethod =>
+      FaceDetectionMethod.fromMlVersion(mlVersion);
+  FaceAlignmentMethod get faceAlignmentMethod =>
+      FaceAlignmentMethod.fromMlVersion(mlVersion);
+  FaceEmbeddingMethod get faceEmbeddingMethod =>
+      FaceEmbeddingMethod.fromMlVersion(mlVersion);
+
+  const FaceMlResult({
+    required this.fileId,
+    required this.faces,
+    required this.mlVersion,
+    required this.errorOccured,
+    required this.onlyThumbnailUsed,
+    required this.faceDetectionImageSize,
+    this.faceAlignmentImageSize,
+  });
+
+  Map<String, dynamic> _toJson() => {
+        'fileId': fileId,
+        'faces': faces.map((face) => face.toJson()).toList(),
+        'mlVersion': mlVersion,
+        'errorOccured': errorOccured,
+        'onlyThumbnailUsed': onlyThumbnailUsed,
+        if (faceDetectionImageSize != null)
+          'faceDetectionImageSize': {
+            'width': faceDetectionImageSize!.width,
+            'height': faceDetectionImageSize!.height,
+          },
+        if (faceAlignmentImageSize != null)
+          'faceAlignmentImageSize': {
+            'width': faceAlignmentImageSize!.width,
+            'height': faceAlignmentImageSize!.height,
+          },
+      };
+
+  String toJsonString() => jsonEncode(_toJson());
+
+  static FaceMlResult _fromJson(Map<String, dynamic> json) {
+    return FaceMlResult(
+      fileId: json['fileId'],
+      faces: (json['faces'] as List)
+          .map((item) => FaceResult.fromJson(item as Map<String, dynamic>))
+          .toList(),
+      mlVersion: json['mlVersion'],
+      errorOccured: json['errorOccured'] ?? false,
+      onlyThumbnailUsed: json['onlyThumbnailUsed'] ?? false,
+      faceDetectionImageSize: json['faceDetectionImageSize'] == null
+          ? null
+          : Size(
+              json['faceDetectionImageSize']['width'],
+              json['faceDetectionImageSize']['height'],
+            ),
+      faceAlignmentImageSize: json['faceAlignmentImageSize'] == null
+          ? null
+          : Size(
+              json['faceAlignmentImageSize']['width'],
+              json['faceAlignmentImageSize']['height'],
+            ),
+    );
+  }
+
+  static FaceMlResult fromJsonString(String jsonString) {
+    return _fromJson(jsonDecode(jsonString));
+  }
+
+  /// Sets the embeddings of the faces with the given faceIds to [10, 10,..., 10].
+  ///
+  /// Throws an exception if a faceId is not found in the FaceMlResult.
+  void setEmbeddingsToTen(List<String> faceIds) {
+    for (final faceId in faceIds) {
+      final faceIndex = faces.indexWhere((face) => face.faceId == faceId);
+      if (faceIndex == -1) {
+        throw Exception("No face found with faceId $faceId");
+      }
+      for (var i = 0; i < faces[faceIndex].embedding.length; i++) {
+        faces[faceIndex].embedding[i] = 10;
+      }
+    }
+  }
+
+  FaceDetectionRelative getDetectionForFaceId(String faceId) {
+    final faceIndex = faces.indexWhere((face) => face.faceId == faceId);
+    if (faceIndex == -1) {
+      throw Exception("No face found with faceId $faceId");
+    }
+    return faces[faceIndex].detection;
+  }
+}
+
+class FaceMlResultBuilder {
+  int fileId;
+
+  List<FaceResultBuilder> faces = <FaceResultBuilder>[];
+
+  Size? faceDetectionImageSize;
+  Size? faceAlignmentImageSize;
+
+  int mlVersion;
+  bool errorOccured;
+  bool onlyThumbnailUsed;
+
+  FaceMlResultBuilder({
+    this.fileId = -1,
+    this.mlVersion = faceMlVersion,
+    this.errorOccured = false,
+    this.onlyThumbnailUsed = false,
+  });
+
+  FaceMlResultBuilder.fromEnteFile(
+    EnteFile file, {
+    this.mlVersion = faceMlVersion,
+    this.errorOccured = false,
+    this.onlyThumbnailUsed = false,
+  }) : fileId = file.uploadedFileID ?? -1;
+
+  FaceMlResultBuilder.fromEnteFileID(
+    int fileID, {
+    this.mlVersion = faceMlVersion,
+    this.errorOccured = false,
+    this.onlyThumbnailUsed = false,
+  }) : fileId = fileID;
+
+  void addNewlyDetectedFaces(
+    List<FaceDetectionRelative> faceDetections,
+    Size originalSize,
+  ) {
+    faceDetectionImageSize = originalSize;
+    for (var i = 0; i < faceDetections.length; i++) {
+      faces.add(
+        FaceResultBuilder.fromFaceDetection(
+          faceDetections[i],
+          resultBuilder: this,
+        ),
+      );
+    }
+  }
+
+  void addAlignmentResults(
+    List<AlignmentResult> alignmentResults,
+    List<double> blurValues,
+    Size imageSizeUsedForAlignment,
+  ) {
+    if (alignmentResults.length != faces.length) {
+      throw Exception(
+        "The amount of alignment results (${alignmentResults.length}) does not match the number of faces (${faces.length})",
+      );
+    }
+
+    for (var i = 0; i < alignmentResults.length; i++) {
+      faces[i].alignment = alignmentResults[i];
+      faces[i].blurValue = blurValues[i];
+    }
+    faceAlignmentImageSize = imageSizeUsedForAlignment;
+  }
+
+  void addEmbeddingsToExistingFaces(
+    List<Embedding> embeddings,
+  ) {
+    if (embeddings.length != faces.length) {
+      throw Exception(
+        "The amount of embeddings (${embeddings.length}) does not match the number of faces (${faces.length})",
+      );
+    }
+    for (var faceIndex = 0; faceIndex < faces.length; faceIndex++) {
+      faces[faceIndex].embedding = embeddings[faceIndex];
+    }
+  }
+
+  FaceMlResult build() {
+    final faceResults = <FaceResult>[];
+    for (var i = 0; i < faces.length; i++) {
+      faceResults.add(faces[i].build());
+    }
+    return FaceMlResult(
+      fileId: fileId,
+      faces: faceResults,
+      mlVersion: mlVersion,
+      errorOccured: errorOccured,
+      onlyThumbnailUsed: onlyThumbnailUsed,
+      faceDetectionImageSize: faceDetectionImageSize,
+      faceAlignmentImageSize: faceAlignmentImageSize,
+    );
+  }
+
+  FaceMlResult buildNoFaceDetected() {
+    faces = <FaceResultBuilder>[];
+    return build();
+  }
+
+  FaceMlResult buildErrorOccurred() {
+    faces = <FaceResultBuilder>[];
+    errorOccured = true;
+    return build();
+  }
+}
+
+@immutable
+class FaceResult {
+  final FaceDetectionRelative detection;
+  final double blurValue;
+  final AlignmentResult alignment;
+  final Embedding embedding;
+  final int fileId;
+  final String faceId;
+
+  bool get isBlurry => blurValue < kLaplacianThreshold;
+
+  const FaceResult({
+    required this.detection,
+    required this.blurValue,
+    required this.alignment,
+    required this.embedding,
+    required this.fileId,
+    required this.faceId,
+  });
+
+  Map<String, dynamic> toJson() => {
+        'detection': detection.toJson(),
+        'blurValue': blurValue,
+        'alignment': alignment.toJson(),
+        'embedding': embedding,
+        'fileId': fileId,
+        'faceId': faceId,
+      };
+
+  static FaceResult fromJson(Map<String, dynamic> json) {
+    return FaceResult(
+      detection: FaceDetectionRelative.fromJson(json['detection']),
+      blurValue: json['blurValue'],
+      alignment: AlignmentResult.fromJson(json['alignment']),
+      embedding: Embedding.from(json['embedding']),
+      fileId: json['fileId'],
+      faceId: json['faceId'],
+    );
+  }
+}
+
+class FaceResultBuilder {
+  FaceDetectionRelative detection =
+      FaceDetectionRelative.defaultInitialization();
+  double blurValue = 1000;
+  AlignmentResult alignment = AlignmentResult.empty();
+  Embedding embedding = <double>[];
+  int fileId = -1;
+  String faceId = '';
+
+  bool get isBlurry => blurValue < kLaplacianThreshold;
+
+  FaceResultBuilder({
+    required this.fileId,
+    required this.faceId,
+  });
+
+  FaceResultBuilder.fromFaceDetection(
+    FaceDetectionRelative faceDetection, {
+    required FaceMlResultBuilder resultBuilder,
+  }) {
+    fileId = resultBuilder.fileId;
+    faceId = faceDetection.toFaceID(fileID: resultBuilder.fileId);
+    detection = faceDetection;
+  }
+
+  FaceResult build() {
+    assert(detection.allKeypoints[0][0] <= 1);
+    assert(detection.box[0] <= 1);
+    return FaceResult(
+      detection: detection,
+      blurValue: blurValue,
+      alignment: alignment,
+      embedding: embedding,
+      fileId: fileId,
+      faceId: faceId,
+    );
+  }
+}
+
+int _getFileIdFromFaceId(String faceId) {
+  return int.parse(faceId.split("_")[0]);
+}

mobile/lib/services/face_ml/face_ml_version.dart

@@ -0,0 +1,15 @@
+abstract class VersionedMethod {
+  final String method;
+  final int version;
+
+  VersionedMethod(this.method, [this.version = 0]);
+
+  const VersionedMethod.empty()
+      : method = 'Empty method',
+        version = 0;
+
+  Map<String, dynamic> toJson() => {
+        'method': method,
+        'version': version,
+      };
+}

mobile/lib/services/face_ml/face_search_service.dart

@@ -0,0 +1,120 @@
+import "dart:typed_data";
+
+import "package:logging/logging.dart";
+import "package:photos/db/files_db.dart";
+import "package:photos/db/ml_data_db.dart";
+import "package:photos/models/file/file.dart";
+import 'package:photos/utils/image_ml_isolate.dart';
+import "package:photos/utils/thumbnail_util.dart";
+
+class FaceSearchService {
+  final _logger = Logger("FaceSearchService");
+
+  final _mlDatabase = MlDataDB.instance;
+  final _filesDatabase = FilesDB.instance;
+
+  // singleton pattern
+  FaceSearchService._privateConstructor();
+  static final instance = FaceSearchService._privateConstructor();
+  factory FaceSearchService() => instance;
+
+  /// Returns the personIDs of all clustered people in the database.
+  Future<List<int>> getAllPeople() async {
+    final peopleIds = await _mlDatabase.getAllClusterIds();
+    return peopleIds;
+  }
+
+  /// Returns the thumbnail associated with a given personId.
+  Future<Uint8List?> getPersonThumbnail(int personID) async {
+    // get the cluster associated with the personID
+    final cluster = await _mlDatabase.getClusterResult(personID);
+    if (cluster == null) {
+      _logger.warning(
+        "No cluster found for personID $personID, unable to get thumbnail.",
+      );
+      return null;
+    }
+
+    // get the faceID and fileID you want to use to generate the thumbnail
+    final String thumbnailFaceID = cluster.thumbnailFaceId;
+    final int thumbnailFileID = cluster.thumbnailFileId;
+
+    // get the full file thumbnail
+    final EnteFile enteFile = await _filesDatabase
+        .getFilesFromIDs([thumbnailFileID]).then((value) => value.values.first);
+    final Uint8List? fileThumbnail = await getThumbnail(enteFile);
+    if (fileThumbnail == null) {
+      _logger.warning(
+        "No full file thumbnail found for thumbnail faceID $thumbnailFaceID, unable to get thumbnail.",
+      );
+      return null;
+    }
+
+    // get the face detection for the thumbnail
+    final thumbnailMlResult =
+        await _mlDatabase.getFaceMlResult(thumbnailFileID);
+    if (thumbnailMlResult == null) {
+      _logger.warning(
+        "No face ml result found for thumbnail faceID $thumbnailFaceID, unable to get thumbnail.",
+      );
+      return null;
+    }
+    final detection = thumbnailMlResult.getDetectionForFaceId(thumbnailFaceID);
+
+    // create the thumbnail from the full file thumbnail and the face detection
+    Uint8List faceThumbnail;
+    try {
+      faceThumbnail = await ImageMlIsolate.instance.generateFaceThumbnail(
+        fileThumbnail,
+        detection,
+      );
+    } catch (e, s) {
+      _logger.warning(
+        "Unable to generate face thumbnail for thumbnail faceID $thumbnailFaceID, unable to get thumbnail.",
+        e,
+        s,
+      );
+      return null;
+    }
+
+    return faceThumbnail;
+  }
+
+  /// Returns all files associated with a given personId.
+  Future<List<EnteFile>> getFilesForPerson(int personID) async {
+    final fileIDs = await _mlDatabase.getClusterFileIds(personID);
+
+    final Map<int, EnteFile> files =
+        await _filesDatabase.getFilesFromIDs(fileIDs);
+    return files.values.toList();
+  }
+
+  Future<List<EnteFile>> getFilesForIntersectOfPeople(
+    List<int> personIDs,
+  ) async {
+    if (personIDs.length <= 1) {
+      _logger
+          .warning('Cannot get intersection of files for less than 2 people');
+      return <EnteFile>[];
+    }
+
+    final Set<int> fileIDsFirstCluster = await _mlDatabase
+        .getClusterFileIds(personIDs.first)
+        .then((value) => value.toSet());
+    for (final personID in personIDs.sublist(1)) {
+      final fileIDsSingleCluster =
+          await _mlDatabase.getClusterFileIds(personID);
+      fileIDsFirstCluster.retainAll(fileIDsSingleCluster);
+
+      // Early termination if intersection is empty
+      if (fileIDsFirstCluster.isEmpty) {
+        return <EnteFile>[];
+      }
+    }
+
+    final Map<int, EnteFile> files =
+        await _filesDatabase.getFilesFromIDs(fileIDsFirstCluster.toList());
+
+    return files.values.toList();
+  }
+}

mobile/lib/services/face_ml/feedback/cluster_feedback.dart

@@ -0,0 +1,464 @@
+import 'dart:developer' as dev;
+import "dart:math" show Random;
+import "dart:typed_data";
+
+import "package:flutter/foundation.dart";
+import "package:logging/logging.dart";
+import "package:photos/core/event_bus.dart";
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/extensions/stop_watch.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/generated/protos/ente/common/vector.pb.dart";
+import "package:photos/models/file/file.dart";
+import "package:photos/services/face_ml/face_clustering/cosine_distance.dart";
+import "package:photos/services/search_service.dart";
+
+class ClusterFeedbackService {
+  final Logger _logger = Logger("ClusterFeedbackService");
+  ClusterFeedbackService._privateConstructor();
+
+  static final ClusterFeedbackService instance =
+      ClusterFeedbackService._privateConstructor();
+
+  /// Returns a map of person's clusterID to map of closest clusterID to with disstance
+  Future<Map<int, List<(int, double)>>> getSuggestionsUsingMean(
+    Person p, {
+    double maxClusterDistance = 0.4,
+  }) async {
+    // Get all the cluster data
+    final faceMlDb = FaceMLDataDB.instance;
+
+    final allClusterIdsToCountMap = (await faceMlDb.clusterIdToFaceCount());
+    final ignoredClusters = await faceMlDb.getPersonIgnoredClusters(p.remoteID);
+    final personClusters = await faceMlDb.getPersonClusterIDs(p.remoteID);
+    dev.log(
+      'existing clusters for ${p.attr.name} are $personClusters',
+      name: "ClusterFeedbackService",
+    );
+
+    // Get and update the cluster summary to get the avg (centroid) and count
+    final EnteWatch watch = EnteWatch("ClusterFeedbackService")..start();
+    final Map<int, List<double>> clusterAvg = await _getUpdateClusterAvg(
+      allClusterIdsToCountMap,
+      ignoredClusters,
+    );
+    watch.log('computed avg for ${clusterAvg.length} clusters');
+
+    // Find the actual closest clusters for the person
+    final Map<int, List<(int, double)>> suggestions = _calcSuggestionsMean(
+      clusterAvg,
+      personClusters,
+      ignoredClusters,
+      maxClusterDistance,
+    );
+
+    // log suggestions
+    for (final entry in suggestions.entries) {
+      dev.log(
+        ' ${entry.value.length} suggestion for ${p.attr.name} for cluster ID ${entry.key} are  suggestions ${entry.value}}',
+        name: "ClusterFeedbackService",
+      );
+    }
+    return suggestions;
+  }
+
+  Future<List<int>> getSuggestionsUsingMedian(
+    Person p, {
+    int sampleSize = 50,
+    double maxMedianDistance = 0.65,
+    double goodMedianDistance = 0.55,
+    double maxMeanDistance = 0.65,
+    double goodMeanDistance = 0.4,
+  }) async {
+    // Get all the cluster data
+    final faceMlDb = FaceMLDataDB.instance;
+    // final Map<int, List<(int, double)>> suggestions = {};
+    final allClusterIdsToCountMap = (await faceMlDb.clusterIdToFaceCount());
+    final ignoredClusters = await faceMlDb.getPersonIgnoredClusters(p.remoteID);
+    final personClusters = await faceMlDb.getPersonClusterIDs(p.remoteID);
+    dev.log(
+      'existing clusters for ${p.attr.name} are $personClusters',
+      name: "ClusterFeedbackService",
+    );
+
+    // Get and update the cluster summary to get the avg (centroid) and count
+    final EnteWatch watch = EnteWatch("ClusterFeedbackService")..start();
+    final Map<int, List<double>> clusterAvg = await _getUpdateClusterAvg(
+      allClusterIdsToCountMap,
+      ignoredClusters,
+    );
+    watch.log('computed avg for ${clusterAvg.length} clusters');
+
+    // Find the other cluster candidates based on the mean
+    final Map<int, List<(int, double)>> suggestionsMean = _calcSuggestionsMean(
+      clusterAvg,
+      personClusters,
+      ignoredClusters,
+      goodMeanDistance,
+    );
+    if (suggestionsMean.isNotEmpty) {
+      final List<(int, double)> suggestClusterIds = [];
+      for (final List<(int, double)> suggestion in suggestionsMean.values) {
+        suggestClusterIds.addAll(suggestion);
+      }
+      suggestClusterIds.sort(
+        (a, b) => allClusterIdsToCountMap[b.$1]!
+            .compareTo(allClusterIdsToCountMap[a.$1]!),
+      );
+      final suggestClusterIdsSizes = suggestClusterIds
+          .map((e) => allClusterIdsToCountMap[e.$1]!)
+          .toList(growable: false);
+      final suggestClusterIdsDistances =
+          suggestClusterIds.map((e) => e.$2).toList(growable: false);
+      _logger.info(
+        "Already found good suggestions using mean: $suggestClusterIds, with sizes $suggestClusterIdsSizes and distances $suggestClusterIdsDistances",
+      );
+      return suggestClusterIds.map((e) => e.$1).toList(growable: false);
+    }
+
+    // Find the other cluster candidates based on the median
+    final Map<int, List<(int, double)>> moreSuggestionsMean =
+        _calcSuggestionsMean(
+      clusterAvg,
+      personClusters,
+      ignoredClusters,
+      maxMeanDistance,
+    );
+    if (moreSuggestionsMean.isEmpty) {
+      _logger
+          .info("No suggestions found using mean, even with higher threshold");
+      return <int>[];
+    }
+
+    final List<(int, double)> temp = [];
+    for (final List<(int, double)> suggestion in moreSuggestionsMean.values) {
+      temp.addAll(suggestion);
+    }
+    temp.sort((a, b) => a.$2.compareTo(b.$2));
+    final otherClusterIdsCandidates = temp
+        .map(
+          (e) => e.$1,
+        )
+        .toList(growable: false);
+    _logger.info(
+      "Found potential suggestions from loose mean for median test: $otherClusterIdsCandidates",
+    );
+
+    watch.logAndReset("Starting median test");
+    // Take the embeddings from the person's clusters in one big list and sample from it
+    final List<Uint8List> personEmbeddingsProto = [];
+    for (final clusterID in personClusters) {
+      final Iterable<Uint8List> embedings =
+          await FaceMLDataDB.instance.getFaceEmbeddingsForCluster(clusterID);
+      personEmbeddingsProto.addAll(embedings);
+    }
+    final List<Uint8List> sampledEmbeddingsProto =
+        _randomSampleWithoutReplacement(
+      personEmbeddingsProto,
+      sampleSize,
+    );
+    final List<List<double>> sampledEmbeddings = sampledEmbeddingsProto
+        .map((embedding) => EVector.fromBuffer(embedding).values)
+        .toList(growable: false);
+
+    // Find the actual closest clusters for the person using median
+    final List<(int, double)> suggestionsMedian = [];
+    final List<(int, double)> greatSuggestionsMedian = [];
+    double minMedianDistance = maxMedianDistance;
+    for (final otherClusterId in otherClusterIdsCandidates) {
+      final Iterable<Uint8List> otherEmbeddingsProto =
+          await FaceMLDataDB.instance.getFaceEmbeddingsForCluster(
+        otherClusterId,
+      );
+      final sampledOtherEmbeddingsProto = _randomSampleWithoutReplacement(
+        otherEmbeddingsProto,
+        sampleSize,
+      );
+      final List<List<double>> sampledOtherEmbeddings =
+          sampledOtherEmbeddingsProto
+              .map((embedding) => EVector.fromBuffer(embedding).values)
+              .toList(growable: false);
+
+      // Calculate distances and find the median
+      final List<double> distances = [];
+      for (final otherEmbedding in sampledOtherEmbeddings) {
+        for (final embedding in sampledEmbeddings) {
+          distances.add(cosineDistForNormVectors(embedding, otherEmbedding));
+        }
+      }
+      distances.sort();
+      final double medianDistance = distances[distances.length ~/ 2];
+      if (medianDistance < minMedianDistance) {
+        suggestionsMedian.add((otherClusterId, medianDistance));
+        minMedianDistance = medianDistance;
+        if (medianDistance < goodMedianDistance) {
+          greatSuggestionsMedian.add((otherClusterId, medianDistance));
+          break;
+        }
+      }
+    }
+    watch.log("Finished median test");
+    if (suggestionsMedian.isEmpty) {
+      _logger.info("No suggestions found using median");
+      return <int>[];
+    } else {
+      _logger.info("Found suggestions using median: $suggestionsMedian");
+    }
+
+    final List<int> finalSuggestionsMedian = suggestionsMedian
+        .map(((e) => e.$1))
+        .toList(growable: false)
+        .reversed
+        .toList(growable: false);
+
+    if (greatSuggestionsMedian.isNotEmpty) {
+      _logger.info(
+        "Found great suggestion using median: $greatSuggestionsMedian",
+      );
+      // // Return the largest size cluster by using allClusterIdsToCountMap
+      // final List<int> greatSuggestionsMedianClusterIds =
+      //     greatSuggestionsMedian.map((e) => e.$1).toList(growable: false);
+      // greatSuggestionsMedianClusterIds.sort(
+      //   (a, b) =>
+      //       allClusterIdsToCountMap[b]!.compareTo(allClusterIdsToCountMap[a]!),
+      // );
+
+      // return [greatSuggestionsMedian.last.$1, ...finalSuggestionsMedian];
+    }
+
+    return finalSuggestionsMedian;
+  }
+
+  Future<List<(int, List<EnteFile>)>> getClusterFilesForPersonID(
+    Person person,
+  ) async {
+    _logger.info(
+      'getClusterFilesForPersonID ${kDebugMode ? person.attr.name : person.remoteID}',
+    );
+
+    // Get the suggestions for the person using only centroids
+    // final Map<int, List<(int, double)>> suggestions =
+    //     await getSuggestionsUsingMean(person);
+    // final Set<int> suggestClusterIds = {};
+    // for (final List<(int, double)> suggestion in suggestions.values) {
+    //   for (final clusterNeighbors in suggestion) {
+    //     suggestClusterIds.add(clusterNeighbors.$1);
+    //   }
+    // }
+
+    try {
+      // Get the suggestions for the person using centroids and median
+      final List<int> suggestClusterIds =
+          await getSuggestionsUsingMedian(person);
+
+      // Get the files for the suggestions
+      final Map<int, Set<int>> fileIdToClusterID = await FaceMLDataDB.instance
+          .getFileIdToClusterIDSetForCluster(suggestClusterIds.toSet());
+      final Map<int, List<EnteFile>> clusterIDToFiles = {};
+      final allFiles = await SearchService.instance.getAllFiles();
+      for (final f in allFiles) {
+        if (!fileIdToClusterID.containsKey(f.uploadedFileID ?? -1)) {
+          continue;
+        }
+        final cluserIds = fileIdToClusterID[f.uploadedFileID ?? -1]!;
+        for (final cluster in cluserIds) {
+          if (clusterIDToFiles.containsKey(cluster)) {
+            clusterIDToFiles[cluster]!.add(f);
+          } else {
+            clusterIDToFiles[cluster] = [f];
+          }
+        }
+      }
+
+      final List<(int, List<EnteFile>)> clusterIdAndFiles = [];
+      for (final clusterId in suggestClusterIds) {
+        if (clusterIDToFiles.containsKey(clusterId)) {
+          clusterIdAndFiles.add(
+            (clusterId, clusterIDToFiles[clusterId]!),
+          );
+        }
+      }
+
+      return clusterIdAndFiles;
+    } catch (e, s) {
+      _logger.severe("Error in getClusterFilesForPersonID", e, s);
+      rethrow;
+    }
+  }
+
+  Future<void> removePersonFromFiles(List<EnteFile> files, Person p) {
+    return FaceMLDataDB.instance.removePersonFromFiles(files, p);
+  }
+
+  Future<bool> checkAndDoAutomaticMerges(Person p) async {
+    final faceMlDb = FaceMLDataDB.instance;
+    final allClusterIdsToCountMap = (await faceMlDb.clusterIdToFaceCount());
+    final ignoredClusters = await faceMlDb.getPersonIgnoredClusters(p.remoteID);
+    final personClusters = await faceMlDb.getPersonClusterIDs(p.remoteID);
+    dev.log(
+      'existing clusters for ${p.attr.name} are $personClusters',
+      name: "ClusterFeedbackService",
+    );
+
+    // Get and update the cluster summary to get the avg (centroid) and count
+    final EnteWatch watch = EnteWatch("ClusterFeedbackService")..start();
+    final Map<int, List<double>> clusterAvg = await _getUpdateClusterAvg(
+      allClusterIdsToCountMap,
+      ignoredClusters,
+    );
+    watch.log('computed avg for ${clusterAvg.length} clusters');
+
+    // Find the actual closest clusters for the person
+    final Map<int, List<(int, double)>> suggestions = _calcSuggestionsMean(
+      clusterAvg,
+      personClusters,
+      ignoredClusters,
+      0.3,
+    );
+
+    if (suggestions.isEmpty) {
+      dev.log(
+        'No automatic merge suggestions for ${p.attr.name}',
+        name: "ClusterFeedbackService",
+      );
+      return false;
+    }
+
+    // log suggestions
+    for (final entry in suggestions.entries) {
+      dev.log(
+        ' ${entry.value.length} suggestion for ${p.attr.name} for cluster ID ${entry.key} are  suggestions ${entry.value}}',
+        name: "ClusterFeedbackService",
+      );
+    }
+
+    for (final suggestionsPerCluster in suggestions.values) {
+      for (final suggestion in suggestionsPerCluster) {
+        final clusterID = suggestion.$1;
+        await faceMlDb.assignClusterToPerson(
+          personID: p.remoteID,
+          clusterID: clusterID,
+        );
+      }
+    }
+
+    Bus.instance.fire(PeopleChangedEvent());
+
+    return true;
+  }
+
+  Future<Map<int, List<double>>> _getUpdateClusterAvg(
+    Map<int, int> allClusterIdsToCountMap,
+    Set<int> ignoredClusters,
+  ) async {
+    final faceMlDb = FaceMLDataDB.instance;
+
+    final Map<int, (Uint8List, int)> clusterToSummary =
+        await faceMlDb.clusterSummaryAll();
+    final Map<int, (Uint8List, int)> updatesForClusterSummary = {};
+
+    final Map<int, List<double>> clusterAvg = {};
+
+    final allClusterIds = allClusterIdsToCountMap.keys;
+    for (final clusterID in allClusterIds) {
+      if (ignoredClusters.contains(clusterID)) {
+        continue;
+      }
+      late List<double> avg;
+      if (clusterToSummary[clusterID]?.$2 ==
+          allClusterIdsToCountMap[clusterID]) {
+        avg = EVector.fromBuffer(clusterToSummary[clusterID]!.$1).values;
+      } else {
+        final Iterable<Uint8List> embedings =
+            await FaceMLDataDB.instance.getFaceEmbeddingsForCluster(clusterID);
+        final List<double> sum = List.filled(192, 0);
+        for (final embedding in embedings) {
+          final data = EVector.fromBuffer(embedding).values;
+          for (int i = 0; i < sum.length; i++) {
+            sum[i] += data[i];
+          }
+        }
+        avg = sum.map((e) => e / embedings.length).toList();
+        final avgEmbeedingBuffer = EVector(values: avg).writeToBuffer();
+        updatesForClusterSummary[clusterID] =
+            (avgEmbeedingBuffer, embedings.length);
+      }
+      clusterAvg[clusterID] = avg;
+    }
+    if (updatesForClusterSummary.isNotEmpty) {
+      await faceMlDb.clusterSummaryUpdate(updatesForClusterSummary);
+    }
+
+    return clusterAvg;
+  }
+
+  Map<int, List<(int, double)>> _calcSuggestionsMean(
+    Map<int, List<double>> clusterAvg,
+    Set<int> personClusters,
+    Set<int> ignoredClusters,
+    double maxClusterDistance,
+  ) {
+    final Map<int, List<(int, double)>> suggestions = {};
+    for (final otherClusterID in clusterAvg.keys) {
+      // ignore the cluster that belong to the person or is ignored
+      if (personClusters.contains(otherClusterID) ||
+          ignoredClusters.contains(otherClusterID)) {
+        continue;
+      }
+      final otherAvg = clusterAvg[otherClusterID]!;
+      int? nearestPersonCluster;
+      double? minDistance;
+      for (final personCluster in personClusters) {
+        final avg = clusterAvg[personCluster]!;
+        final distance = cosineDistForNormVectors(avg, otherAvg);
+        if (distance < maxClusterDistance) {
+          if (minDistance == null || distance < minDistance) {
+            minDistance = distance;
+            nearestPersonCluster = personCluster;
+          }
+        }
+      }
+      if (nearestPersonCluster != null && minDistance != null) {
+        suggestions
+            .putIfAbsent(nearestPersonCluster, () => [])
+            .add((otherClusterID, minDistance));
+      }
+    }
+    for (final entry in suggestions.entries) {
+      entry.value.sort((a, b) => a.$1.compareTo(b.$1));
+    }
+
+    return suggestions;
+  }
+
+  List<T> _randomSampleWithoutReplacement<T>(
+    Iterable<T> embeddings,
+    int sampleSize,
+  ) {
+    final random = Random();
+
+    if (sampleSize >= embeddings.length) {
+      return embeddings.toList();
+    }
+
+    // If sampleSize is more than half the list size, shuffle and take first sampleSize elements
+    if (sampleSize > embeddings.length / 2) {
+      final List<T> shuffled = List<T>.from(embeddings)..shuffle(random);
+      return shuffled.take(sampleSize).toList(growable: false);
+    }
+
+    // Otherwise, use the set-based method for efficiency
+    final selectedIndices = <int>{};
+    final sampledEmbeddings = <T>[];
+    while (sampledEmbeddings.length < sampleSize) {
+      final int index = random.nextInt(embeddings.length);
+      if (!selectedIndices.contains(index)) {
+        selectedIndices.add(index);
+        sampledEmbeddings.add(embeddings.elementAt(index));
+      }
+    }
+
+    return sampledEmbeddings;
+  }
+}

mobile/lib/services/face_ml/model_file.dart

@@ -0,0 +1,11 @@
+mixin ModelFile {
+  static const String faceDetectionBackWeb =
+      'assets/models/blazeface/blazeface_back_ente_web.tflite';
+  // TODO: which of the two mobilefacenet model should I use now??
+  // static const String faceEmbeddingEnte =
+  // 'assets/models/mobilefacenet/mobilefacenet_ente_web.tflite';
+  static const String faceEmbeddingEnte =
+      'assets/models/mobilefacenet/mobilefacenet_unq_TF211.tflite';
+  static const String yoloV5FaceS640x640DynamicBatchonnx =
+      'assets/models/yolov5face/yolov5s_face_640_640_dynamic.onnx';
+}

mobile/lib/services/search_service.dart

@@ -11,6 +11,8 @@ import 'package:photos/data/years.dart';
 import 'package:photos/db/files_db.dart';
 import 'package:photos/events/local_photos_updated_event.dart';
 import "package:photos/extensions/string_ext.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
 import "package:photos/models/api/collection/user.dart";
 import 'package:photos/models/collection/collection.dart';
 import 'package:photos/models/collection/collection_items.dart';
@@ -22,6 +24,7 @@ import "package:photos/models/location/location.dart";
 import "package:photos/models/location_tag/location_tag.dart";
 import 'package:photos/models/search/album_search_result.dart';
 import 'package:photos/models/search/generic_search_result.dart';
+import "package:photos/models/search/search_constants.dart";
 import "package:photos/models/search/search_types.dart";
 import 'package:photos/services/collections_service.dart';
 import "package:photos/services/location_service.dart";
@@ -29,6 +32,8 @@ import 'package:photos/services/machine_learning/semantic_search/semantic_search
 import "package:photos/states/location_screen_state.dart";
 import "package:photos/ui/viewer/location/add_location_sheet.dart";
 import "package:photos/ui/viewer/location/location_screen.dart";
+import "package:photos/ui/viewer/people/cluster_page.dart";
+import "package:photos/ui/viewer/people/people_page.dart";
 import 'package:photos/utils/date_time_util.dart';
 import "package:photos/utils/navigation_util.dart";
 import 'package:tuple/tuple.dart';
@@ -704,6 +709,146 @@ class SearchService {
     return searchResults;
   }
 
+  Future<Map<int, List<EnteFile>>> getClusterFilesForPersonID(
+    String personID,
+  ) async {
+    _logger.info('getClusterFilesForPersonID $personID');
+    final Map<int, Set<int>> fileIdToClusterID =
+        await FaceMLDataDB.instance.getFileIdToClusterIDSet(personID);
+    _logger.info('faceDbDone getClusterFilesForPersonID $personID');
+    final Map<int, List<EnteFile>> clusterIDToFiles = {};
+    final allFiles = await getAllFiles();
+    for (final f in allFiles) {
+      if (!fileIdToClusterID.containsKey(f.uploadedFileID ?? -1)) {
+        continue;
+      }
+      final cluserIds = fileIdToClusterID[f.uploadedFileID ?? -1]!;
+      for (final cluster in cluserIds) {
+        if (clusterIDToFiles.containsKey(cluster)) {
+          clusterIDToFiles[cluster]!.add(f);
+        } else {
+          clusterIDToFiles[cluster] = [f];
+        }
+      }
+    }
+    _logger.info('done getClusterFilesForPersonID $personID');
+    return clusterIDToFiles;
+  }
+
+  Future<List<GenericSearchResult>> getAllFace(int? limit) async {
+    debugPrint("getting faces");
+    final Map<int, Set<int>> fileIdToClusterID =
+        await FaceMLDataDB.instance.getFileIdToClusterIds();
+    final (clusterIDToPerson, personIdToPerson) =
+        await FaceMLDataDB.instance.getClusterIdToPerson();
+
+    debugPrint("building result");
+    final List<GenericSearchResult> facesResult = [];
+    final Map<int, List<EnteFile>> clusterIdToFiles = {};
+    final Map<String, List<EnteFile>> personIdToFiles = {};
+    final allFiles = await getAllFiles();
+    for (final f in allFiles) {
+      if (!fileIdToClusterID.containsKey(f.uploadedFileID ?? -1)) {
+        continue;
+      }
+      final cluserIds = fileIdToClusterID[f.uploadedFileID ?? -1]!;
+      for (final cluster in cluserIds) {
+        final Person? p = clusterIDToPerson[cluster];
+        if (p != null) {
+          if (personIdToFiles.containsKey(p.remoteID)) {
+            personIdToFiles[p.remoteID]!.add(f);
+          } else {
+            personIdToFiles[p.remoteID] = [f];
+          }
+        } else {
+          if (clusterIdToFiles.containsKey(cluster)) {
+            clusterIdToFiles[cluster]!.add(f);
+          } else {
+            clusterIdToFiles[cluster] = [f];
+          }
+        }
+      }
+    }
+    // get sorted personId by files count
+    final sortedPersonIds = personIdToFiles.keys.toList()
+      ..sort(
+        (a, b) => personIdToFiles[b]!.length.compareTo(
+              personIdToFiles[a]!.length,
+            ),
+      );
+    for (final personID in sortedPersonIds) {
+      final files = personIdToFiles[personID]!;
+      if (files.isEmpty) {
+        continue;
+      }
+      final Person p = personIdToPerson[personID]!;
+      facesResult.add(
+        GenericSearchResult(
+          ResultType.faces,
+          p.attr.name,
+          files,
+          params: {
+            kPersonParamID: personID,
+            kFileID: files.first.uploadedFileID,
+          },
+          onResultTap: (ctx) {
+            routeToPage(
+              ctx,
+              PeoplePage(
+                tagPrefix: "${ResultType.faces.toString()}_${p.attr.name}",
+                person: p,
+              ),
+            );
+          },
+        ),
+      );
+    }
+    final sortedClusterIds = clusterIdToFiles.keys.toList()
+      ..sort(
+        (a, b) =>
+            clusterIdToFiles[b]!.length.compareTo(clusterIdToFiles[a]!.length),
+      );
+
+    for (final clusterId in sortedClusterIds) {
+      final files = clusterIdToFiles[clusterId]!;
+      // final String clusterName = "ID:$clusterId,  ${files.length}";
+      final String clusterName = "${files.length}";
+      final Person? p = clusterIDToPerson[clusterId];
+      if (p != null) {
+        throw Exception("Person should  be null");
+      }
+      if (files.length < 3) {
+        continue;
+      }
+      facesResult.add(
+        GenericSearchResult(
+          ResultType.faces,
+          clusterName,
+          files,
+          params: {
+            kClusterParamId: clusterId,
+            kFileID: files.first.uploadedFileID,
+          },
+          onResultTap: (ctx) {
+            routeToPage(
+              ctx,
+              ClusterPage(
+                files,
+                tagPrefix: "${ResultType.faces.toString()}_$clusterName",
+                cluserID: clusterId,
+              ),
+            );
+          },
+        ),
+      );
+    }
+    if (limit != null) {
+      return facesResult.sublist(0, min(limit, facesResult.length));
+    } else {
+      return facesResult;
+    }
+  }
+
   Future<List<GenericSearchResult>> getAllLocationTags(int? limit) async {
     try {
       final Map<LocalEntity<LocationTag>, List<EnteFile>> tagToItemsMap = {};

mobile/lib/states/all_sections_examples_state.dart

@@ -6,6 +6,7 @@ import "package:logging/logging.dart";
 import "package:photos/core/constants.dart";
 import "package:photos/core/event_bus.dart";
 import "package:photos/events/files_updated_event.dart";
+import "package:photos/events/people_changed_event.dart";
 import "package:photos/events/tab_changed_event.dart";
 import "package:photos/models/search/search_result.dart";
 import "package:photos/models/search/search_types.dart";
@@ -31,6 +32,7 @@ class _AllSectionsExamplesProviderState
   Future<List<List<SearchResult>>> allSectionsExamplesFuture = Future.value([]);
 
   late StreamSubscription<FilesUpdatedEvent> _filesUpdatedEvent;
+  late StreamSubscription<PeopleChangedEvent> _onPeopleChangedEvent;
   late StreamSubscription<TabChangedEvent> _tabChangeEvent;
   bool hasPendingUpdate = false;
   bool isOnSearchTab = false;
@@ -46,16 +48,11 @@ class _AllSectionsExamplesProviderState
     super.initState();
     //add all common events for all search sections to reload to here.
     _filesUpdatedEvent = Bus.instance.on<FilesUpdatedEvent>().listen((event) {
-      if (!isOnSearchTab) {
-        if (kDebugMode) {
-          _logger.finest('Skip reload till user clicks on search tab');
-        }
-        hasPendingUpdate = true;
-        return;
-      } else {
-        hasPendingUpdate = false;
-        reloadAllSections();
-      }
+      onDataUpdate();
+    });
+    _onPeopleChangedEvent =
+        Bus.instance.on<PeopleChangedEvent>().listen((event) {
+      onDataUpdate();
     });
     _tabChangeEvent = Bus.instance.on<TabChangedEvent>().listen((event) {
       if (event.source == TabChangedEventSource.pageView &&
@@ -72,6 +69,18 @@ class _AllSectionsExamplesProviderState
     reloadAllSections();
   }
 
+  void onDataUpdate() {
+    if (!isOnSearchTab) {
+      if (kDebugMode) {
+        _logger.finest('Skip reload till user clicks on search tab');
+      }
+      hasPendingUpdate = true;
+    } else {
+      hasPendingUpdate = false;
+      reloadAllSections();
+    }
+  }
+
   void reloadAllSections() {
     _logger.info('queue reload all sections');
     _debouncer.run(() async {
@@ -79,22 +88,28 @@ class _AllSectionsExamplesProviderState
         _logger.info("'_debounceTimer: reloading all sections in search tab");
         final allSectionsExamples = <Future<List<SearchResult>>>[];
         for (SectionType sectionType in SectionType.values) {
-          if (sectionType == SectionType.face ||
-              sectionType == SectionType.content) {
+          if (sectionType == SectionType.content) {
             continue;
           }
           allSectionsExamples.add(
             sectionType.getData(context, limit: kSearchSectionLimit),
           );
         }
-        allSectionsExamplesFuture =
-            Future.wait<List<SearchResult>>(allSectionsExamples);
+        try {
+          allSectionsExamplesFuture = Future.wait<List<SearchResult>>(
+            allSectionsExamples,
+            eagerError: false,
+          );
+        } catch (e) {
+          _logger.severe("Error reloading all sections: $e");
+        }
       });
     });
   }
 
   @override
   void dispose() {
+    _onPeopleChangedEvent.cancel();
     _filesUpdatedEvent.cancel();
     _tabChangeEvent.cancel();
     _debouncer.cancelDebounce();

mobile/lib/ui/components/bottom_action_bar/bottom_action_bar_widget.dart

@@ -1,5 +1,6 @@
 import 'package:flutter/material.dart';
 import 'package:photos/core/constants.dart';
+import "package:photos/face/model/person.dart";
 import 'package:photos/models/collection/collection.dart';
 import "package:photos/models/gallery_type.dart";
 import 'package:photos/models/selected_files.dart';
@@ -11,6 +12,8 @@ import "package:photos/ui/viewer/actions/file_selection_actions_widget.dart";
 class BottomActionBarWidget extends StatelessWidget {
   final GalleryType galleryType;
   final Collection? collection;
+  final Person? person;
+  final int? clusterID;
   final SelectedFiles selectedFiles;
   final VoidCallback? onCancel;
   final Color? backgroundColor;
@@ -19,6 +22,8 @@ class BottomActionBarWidget extends StatelessWidget {
     required this.galleryType,
     required this.selectedFiles,
     this.collection,
+    this.person,
+    this.clusterID,
     this.onCancel,
     this.backgroundColor,
     super.key,
@@ -54,6 +59,8 @@ class BottomActionBarWidget extends StatelessWidget {
             galleryType,
             selectedFiles,
             collection: collection,
+            person: person,
+            clusterID: clusterID,
           ),
           const DividerWidget(dividerType: DividerType.bottomBar),
           ActionBarWidget(

mobile/lib/ui/settings/debug/debug_section_widget.dart

@@ -67,7 +67,6 @@ class DebugSectionWidget extends StatelessWidget {
             showShortToast(context, "Done");
           },
         ),
-        sectionOptionSpacing,
       ],
     );
   }

mobile/lib/ui/settings/debug/face_debug_section_widget.dart

@@ -0,0 +1,214 @@
+import "dart:async";
+
+import "package:flutter/foundation.dart";
+import 'package:flutter/material.dart';
+import "package:logging/logging.dart";
+import "package:photos/core/event_bus.dart";
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/extensions/stop_watch.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/services/face_ml/face_ml_service.dart";
+import "package:photos/services/face_ml/feedback/cluster_feedback.dart";
+import 'package:photos/theme/ente_theme.dart';
+import 'package:photos/ui/components/captioned_text_widget.dart';
+import 'package:photos/ui/components/expandable_menu_item_widget.dart';
+import 'package:photos/ui/components/menu_item_widget/menu_item_widget.dart';
+import 'package:photos/ui/settings/common_settings.dart';
+import "package:photos/utils/dialog_util.dart";
+import "package:photos/utils/local_settings.dart";
+import 'package:photos/utils/toast_util.dart';
+
+class FaceDebugSectionWidget extends StatefulWidget {
+  const FaceDebugSectionWidget({Key? key}) : super(key: key);
+
+  @override
+  State<FaceDebugSectionWidget> createState() => _FaceDebugSectionWidgetState();
+}
+
+class _FaceDebugSectionWidgetState extends State<FaceDebugSectionWidget> {
+  Timer? _timer;
+  @override
+  void initState() {
+    super.initState();
+    _timer = Timer.periodic(const Duration(seconds: 5), (timer) {
+      setState(() {
+        // Your state update logic here
+      });
+    });
+  }
+
+  @override
+  void dispose() {
+    _timer?.cancel();
+    super.dispose();
+  }
+
+  @override
+  Widget build(BuildContext context) {
+    return ExpandableMenuItemWidget(
+      title: "Face Beta",
+      selectionOptionsWidget: _getSectionOptions(context),
+      leadingIcon: Icons.bug_report_outlined,
+    );
+  }
+
+  Widget _getSectionOptions(BuildContext context) {
+    final Logger _logger = Logger("FaceDebugSectionWidget");
+    return Column(
+      children: [
+        MenuItemWidget(
+          captionedTextWidget: FutureBuilder<Set<int>>(
+            future: FaceMLDataDB.instance.getIndexedFileIds(),
+            builder: (context, snapshot) {
+              if (snapshot.hasData) {
+                return CaptionedTextWidget(
+                  title: LocalSettings.instance.isFaceIndexingEnabled
+                      ? "Disable Indexing (${snapshot.data!.length})"
+                      : "Enable indexing (${snapshot.data!.length})",
+                );
+              }
+              return const SizedBox.shrink();
+            },
+          ),
+          pressedColor: getEnteColorScheme(context).fillFaint,
+          trailingIcon: Icons.chevron_right_outlined,
+          trailingIconIsMuted: true,
+          onTap: () async {
+            try {
+              final isEnabled =
+                  await LocalSettings.instance.toggleFaceIndexing();
+              if (isEnabled) {
+                FaceMlService.instance.indexAllImages().ignore();
+              } else {
+                FaceMlService.instance.pauseIndexing();
+              }
+              if (mounted) {
+                setState(() {});
+              }
+            } catch (e, s) {
+              _logger.warning('indexing failed ', e, s);
+              await showGenericErrorDialog(context: context, error: e);
+            }
+          },
+        ),
+        MenuItemWidget(
+          captionedTextWidget: const CaptionedTextWidget(
+            title: "Run Clustering",
+          ),
+          pressedColor: getEnteColorScheme(context).fillFaint,
+          trailingIcon: Icons.chevron_right_outlined,
+          trailingIconIsMuted: true,
+          onTap: () async {
+            await FaceMlService.instance.clusterAllImages(minFaceScore: 0.75);
+            Bus.instance.fire(PeopleChangedEvent());
+            showShortToast(context, "Done");
+          },
+        ),
+        sectionOptionSpacing,
+        MenuItemWidget(
+          captionedTextWidget: const CaptionedTextWidget(
+            title: "Reset feedback & labels",
+          ),
+          pressedColor: getEnteColorScheme(context).fillFaint,
+          trailingIcon: Icons.chevron_right_outlined,
+          trailingIconIsMuted: true,
+          onTap: () async {
+            await FaceMLDataDB.instance.resetClusterIDs();
+            await FaceMLDataDB.instance.dropClustersAndPeople();
+            Bus.instance.fire(PeopleChangedEvent());
+            showShortToast(context, "Done");
+          },
+        ),
+        sectionOptionSpacing,
+        MenuItemWidget(
+          captionedTextWidget: const CaptionedTextWidget(
+            title: "Drop embeddings & feedback",
+          ),
+          pressedColor: getEnteColorScheme(context).fillFaint,
+          trailingIcon: Icons.chevron_right_outlined,
+          trailingIconIsMuted: true,
+          onTap: () async {
+            await showChoiceDialog(
+              context,
+              title: "Are you sure?",
+              body:
+                  "You will need to again re-index all the faces. You can drop feedback if you want to label again",
+              firstButtonLabel: "Yes, confirm",
+              firstButtonOnTap: () async {
+                await FaceMLDataDB.instance.dropClustersAndPeople(faces: true);
+                Bus.instance.fire(PeopleChangedEvent());
+                showShortToast(context, "Done");
+              },
+            );
+          },
+        ),
+        if (kDebugMode) sectionOptionSpacing,
+        if (kDebugMode)
+          MenuItemWidget(
+            captionedTextWidget: const CaptionedTextWidget(
+              title: "Pull Embeddings From Local",
+            ),
+            pressedColor: getEnteColorScheme(context).fillFaint,
+            trailingIcon: Icons.chevron_right_outlined,
+            trailingIconIsMuted: true,
+            onTap: () async {
+              try {
+                final List<Person> persons =
+                    await FaceMLDataDB.instance.getPeople();
+                final EnteWatch w = EnteWatch('feedback')..start();
+                for (final Person p in persons) {
+                  await ClusterFeedbackService.instance
+                      .getSuggestionsUsingMean(p);
+                  w.logAndReset('suggestion calculated for ${p.attr.name}');
+                }
+                w.log("done with feedback");
+                showShortToast(context, "done avg");
+                // await FaceMLDataDB.instance.bulkInsertFaces([]);
+                // final EnteWatch watch = EnteWatch("face_time")..start();
+
+                // final results = await downloadZip();
+                // watch.logAndReset('downloaded and de-serialized');
+                // await FaceMLDataDB.instance.bulkInsertFaces(results);
+                // watch.logAndReset('inserted in to db');
+                // showShortToast(context, "Got ${results.length} results");
+              } catch (e, s) {
+                _logger.warning('download failed ', e, s);
+                await showGenericErrorDialog(context: context, error: e);
+              }
+              // _showKeyAttributesDialog(context);
+            },
+          ),
+        if (kDebugMode) sectionOptionSpacing,
+        if (kDebugMode)
+          MenuItemWidget(
+            captionedTextWidget: FutureBuilder<Set<int>>(
+              future: FaceMLDataDB.instance.getIndexedFileIds(),
+              builder: (context, snapshot) {
+                if (snapshot.hasData) {
+                  return CaptionedTextWidget(
+                    title: "Read embeddings for ${snapshot.data!.length} files",
+                  );
+                }
+                return const CaptionedTextWidget(
+                  title: "Loading...",
+                );
+              },
+            ),
+            pressedColor: getEnteColorScheme(context).fillFaint,
+            trailingIcon: Icons.chevron_right_outlined,
+            trailingIconIsMuted: true,
+            onTap: () async {
+              final EnteWatch watch = EnteWatch("read_embeddings")..start();
+              final result = await FaceMLDataDB.instance.getFaceEmbeddingMap();
+              watch.logAndReset('read embeddings ${result.length} ');
+              showShortToast(
+                context,
+                "Done in ${watch.elapsed.inSeconds} secs",
+              );
+            },
+          ),
+      ],
+    );
+  }
+}

mobile/lib/ui/settings_page.dart

@@ -7,7 +7,6 @@ import 'package:photos/core/configuration.dart';
 import 'package:photos/core/event_bus.dart';
 import 'package:photos/events/opened_settings_event.dart';
 import "package:photos/generated/l10n.dart";
-import 'package:photos/services/feature_flag_service.dart';
 import "package:photos/services/storage_bonus_service.dart";
 import 'package:photos/theme/colors.dart';
 import 'package:photos/theme/ente_theme.dart';
@@ -17,7 +16,8 @@ import 'package:photos/ui/settings/about_section_widget.dart';
 import 'package:photos/ui/settings/account_section_widget.dart';
 import 'package:photos/ui/settings/app_version_widget.dart';
 import 'package:photos/ui/settings/backup/backup_section_widget.dart';
-import 'package:photos/ui/settings/debug_section_widget.dart';
+import 'package:photos/ui/settings/debug/debug_section_widget.dart';
+import "package:photos/ui/settings/debug/face_debug_section_widget.dart";
 import 'package:photos/ui/settings/general_section_widget.dart';
 import 'package:photos/ui/settings/inherited_settings_state.dart';
 import 'package:photos/ui/settings/security_section_widget.dart';
@@ -52,6 +52,10 @@ class SettingsPage extends StatelessWidget {
     final hasLoggedIn = Configuration.instance.isLoggedIn();
     final enteTextTheme = getEnteTextTheme(context);
     final List<Widget> contents = [];
+    const sectionSpacing = SizedBox(height: 8);
+    if (kDebugMode) {
+      contents.addAll([const FaceDebugSectionWidget(), sectionSpacing]);
+    }
     contents.add(
       GestureDetector(
         onDoubleTap: () {
@@ -81,7 +85,7 @@ class SettingsPage extends StatelessWidget {
         ),
       ),
     );
-    const sectionSpacing = SizedBox(height: 8);
+
     contents.add(const SizedBox(height: 8));
     if (hasLoggedIn) {
       final showStorageBonusBanner =
@@ -139,9 +143,9 @@ class SettingsPage extends StatelessWidget {
       const AboutSectionWidget(),
     ]);
 
-    if (hasLoggedIn &&
-        FeatureFlagService.instance.isInternalUserOrDebugBuild()) {
+    if (hasLoggedIn) {
       contents.addAll([sectionSpacing, const DebugSectionWidget()]);
+      contents.addAll([sectionSpacing, const FaceDebugSectionWidget()]);
     }
     contents.add(const AppVersionWidget());
     contents.add(

mobile/lib/ui/tools/app_lock.dart

@@ -113,6 +113,7 @@ class _AppLockState extends State<AppLock> with WidgetsBindingObserver {
       theme: widget.lightTheme,
       darkTheme: widget.darkTheme,
       locale: widget.locale,
+      debugShowCheckedModeBanner: false,
       supportedLocales: appSupportedLocales,
       localeListResolutionCallback: localResolutionCallBack,
       localizationsDelegates: const [

mobile/lib/ui/viewer/actions/file_selection_actions_widget.dart

@@ -1,10 +1,15 @@
 import "dart:async";
 
 import 'package:fast_base58/fast_base58.dart';
+import "package:flutter/cupertino.dart";
 import 'package:flutter/material.dart';
 import 'package:flutter/services.dart';
 import "package:modal_bottom_sheet/modal_bottom_sheet.dart";
 import 'package:photos/core/configuration.dart';
+import "package:photos/core/event_bus.dart";
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
 import "package:photos/generated/l10n.dart";
 import 'package:photos/models/collection/collection.dart';
 import 'package:photos/models/device_collection.dart';
@@ -15,6 +20,7 @@ import 'package:photos/models/gallery_type.dart';
 import "package:photos/models/metadata/common_keys.dart";
 import 'package:photos/models/selected_files.dart';
 import 'package:photos/services/collections_service.dart';
+import "package:photos/services/face_ml/feedback/cluster_feedback.dart";
 import 'package:photos/services/hidden_service.dart';
 import "package:photos/theme/colors.dart";
 import "package:photos/theme/ente_theme.dart";
@@ -39,12 +45,16 @@ class FileSelectionActionsWidget extends StatefulWidget {
   final Collection? collection;
   final DeviceCollection? deviceCollection;
   final SelectedFiles selectedFiles;
+  final Person? person;
+  final int? clusterID;
 
   const FileSelectionActionsWidget(
     this.type,
     this.selectedFiles, {
     Key? key,
     this.collection,
+    this.person,
+    this.clusterID,
     this.deviceCollection,
   }) : super(key: key);
 
@@ -116,7 +126,24 @@ class _FileSelectionActionsWidgetState
     //and set [shouldShow] to false for items that should not be shown and true
     //for items that should be shown.
     final List<SelectionActionButton> items = [];
-
+    if (widget.type == GalleryType.peopleTag && widget.person != null) {
+      items.add(
+        SelectionActionButton(
+          icon: Icons.remove_circle_outline,
+          labelText: 'Not ${widget.person!.attr.name}?',
+          onTap: anyUploadedFiles ? _onNotpersonClicked : null,
+        ),
+      );
+      if (ownedFilesCount == 1) {
+        items.add(
+          SelectionActionButton(
+            icon: Icons.image_outlined,
+            labelText: 'Use as cover',
+            onTap: anyUploadedFiles ? _setPersonCover : null,
+          ),
+        );
+      }
+    }
     if (widget.type.showCreateLink()) {
       if (_cachedCollectionForSharedLink != null && anyUploadedFiles) {
         items.add(
@@ -374,6 +401,16 @@ class _FileSelectionActionsWidgetState
       ),
     );
 
+    if (widget.type == GalleryType.cluster) {
+      items.add(
+        SelectionActionButton(
+          labelText: 'Remove',
+          icon: CupertinoIcons.minus,
+          onTap: () => showToast(context, 'yet to implement'),
+        ),
+      );
+    }
+
     if (items.isNotEmpty) {
       final scrollController = ScrollController();
       // h4ck: https://github.com/flutter/flutter/issues/57920#issuecomment-893970066
@@ -613,6 +650,59 @@ class _FileSelectionActionsWidgetState
     }
   }
 
+  Future<void> _setPersonCover() async {
+    final EnteFile file = widget.selectedFiles.files.first;
+    final Person newPerson = widget.person!.copyWith(
+      attr: widget.person!.attr
+          .copyWith(avatarFaceId: file.uploadedFileID.toString()),
+    );
+    await FaceMLDataDB.instance.updatePerson(newPerson);
+    widget.selectedFiles.clearAll();
+    if (mounted) {
+      setState(() => {});
+    }
+    Bus.instance.fire(PeopleChangedEvent());
+  }
+
+  Future<void> _onNotpersonClicked() async {
+    final actionResult = await showActionSheet(
+      context: context,
+      buttons: [
+        ButtonWidget(
+          labelText: S.of(context).yesRemove,
+          buttonType: ButtonType.neutral,
+          buttonSize: ButtonSize.large,
+          shouldStickToDarkTheme: true,
+          buttonAction: ButtonAction.first,
+          isInAlert: true,
+        ),
+        ButtonWidget(
+          labelText: S.of(context).cancel,
+          buttonType: ButtonType.secondary,
+          buttonSize: ButtonSize.large,
+          buttonAction: ButtonAction.second,
+          shouldStickToDarkTheme: true,
+          isInAlert: true,
+        ),
+      ],
+      title: "Remove these photos for ${widget.person!.attr.name}?",
+      actionSheetType: ActionSheetType.defaultActionSheet,
+    );
+    if (actionResult?.action != null) {
+      if (actionResult!.action == ButtonAction.first) {
+        await ClusterFeedbackService.instance.removePersonFromFiles(
+          widget.selectedFiles.files.toList(),
+          widget.person!,
+        );
+      }
+      Bus.instance.fire(PeopleChangedEvent());
+    }
+    widget.selectedFiles.clearAll();
+    if (mounted) {
+      setState(() => {});
+    }
+  }
+
   Future<void> _copyLink() async {
     if (_cachedCollectionForSharedLink != null) {
       final String collectionKey = Base58Encode(

mobile/lib/ui/viewer/actions/file_selection_overlay_bar.dart

@@ -1,4 +1,5 @@
 import 'package:flutter/material.dart';
+import "package:photos/face/model/person.dart";
 import 'package:photos/models/collection/collection.dart';
 import 'package:photos/models/gallery_type.dart';
 import 'package:photos/models/selected_files.dart';
@@ -10,12 +11,14 @@ class FileSelectionOverlayBar extends StatefulWidget {
   final SelectedFiles selectedFiles;
   final Collection? collection;
   final Color? backgroundColor;
+  final Person? person;
 
   const FileSelectionOverlayBar(
     this.galleryType,
     this.selectedFiles, {
     this.collection,
     this.backgroundColor,
+    this.person,
     Key? key,
   }) : super(key: key);
 
@@ -65,6 +68,7 @@ class _FileSelectionOverlayBarState extends State<FileSelectionOverlayBar> {
               selectedFiles: widget.selectedFiles,
               galleryType: widget.galleryType,
               collection: widget.collection,
+              person: widget.person,
               onCancel: () {
                 if (widget.selectedFiles.files.isNotEmpty) {
                   widget.selectedFiles.clearAll();

mobile/lib/ui/viewer/file/file_details_widget.dart

@@ -18,9 +18,9 @@ import "package:photos/ui/viewer/file_details/albums_item_widget.dart";
 import 'package:photos/ui/viewer/file_details/backed_up_time_item_widget.dart';
 import "package:photos/ui/viewer/file_details/creation_time_item_widget.dart";
 import 'package:photos/ui/viewer/file_details/exif_item_widgets.dart';
+import "package:photos/ui/viewer/file_details/faces_item_widget.dart";
 import "package:photos/ui/viewer/file_details/file_properties_item_widget.dart";
 import "package:photos/ui/viewer/file_details/location_tags_widget.dart";
-import "package:photos/ui/viewer/file_details/objects_item_widget.dart";
 import "package:photos/utils/exif_util.dart";
 
 class FileDetailsWidget extends StatefulWidget {
@@ -221,7 +221,8 @@ class _FileDetailsWidgetState extends State<FileDetailsWidget> {
 
     if (!UpdateService.instance.isFdroidFlavor()) {
       fileDetailsTiles.addAll([
-        ObjectsItemWidget(file),
+        // ObjectsItemWidget(file),
+        FacesItemWidget(file),
         const FileDetailsDivider(),
       ]);
     }

mobile/lib/ui/viewer/file/zoomable_image.dart

@@ -1,5 +1,5 @@
 import 'dart:async';
-import 'dart:io';
+import 'dart:io' as io;
 
 import 'package:flutter/material.dart';
 import 'package:flutter/widgets.dart';
@@ -198,7 +198,7 @@ class _ZoomableImageState extends State<ZoomableImage>
       _loadingFinalImage = true;
       getFile(
         _photo,
-        isOrigin: Platform.isIOS &&
+        isOrigin: io.Platform.isIOS &&
             _isGIF(), // since on iOS GIFs playback only when origin-files are loaded
       ).then((file) {
         if (file != null && file.existsSync()) {
@@ -240,7 +240,25 @@ class _ZoomableImageState extends State<ZoomableImage>
     }
   }
 
-  void _onFinalImageLoaded(ImageProvider imageProvider) {
+  void _onFinalImageLoaded(ImageProvider imageProvider) async {
+    // // final result = await FaceMlService.instance.analyzeImage(
+    // //   _photo,
+    // //   preferUsingThumbnailForEverything: false,
+    // //   disposeImageIsolateAfterUse: false,
+    // // );
+    // // _logger.info("FaceMlService result: $result");
+    // // _logger.info("Number of faces detected: ${result.faces.length}");
+    // // _logger.info("Box: ${result.faces[0].detection.box}");
+    // // _logger.info("Landmarks: ${result.faces[0].detection.allKeypoints}");
+    // // final embedding = result.faces[0].embedding;
+    // // Calculate the magnitude of the embedding vector
+    // double sum = 0;
+    // for (final double value in embedding) {
+    //   sum += value * value;
+    // }
+    // final magnitude = math.sqrt(sum);
+    // log("Magnitude: $magnitude");
+    // log("Embedding: $embedding");
     if (mounted) {
       precacheImage(imageProvider, context).then((value) async {
         if (mounted) {

mobile/lib/ui/viewer/file_details/face_widget.dart

@@ -0,0 +1,160 @@
+import "dart:developer" show log;
+import "dart:typed_data";
+
+import "package:flutter/material.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/face.dart";
+import "package:photos/face/model/person.dart";
+import 'package:photos/models/file/file.dart';
+import "package:photos/services/search_service.dart";
+import "package:photos/ui/viewer/file/no_thumbnail_widget.dart";
+import "package:photos/ui/viewer/people/cluster_page.dart";
+import "package:photos/ui/viewer/people/people_page.dart";
+import "package:photos/utils/face/face_box_crop.dart";
+import "package:photos/utils/thumbnail_util.dart";
+
+class FaceWidget extends StatelessWidget {
+  final EnteFile file;
+  final Face face;
+  final Person? person;
+  final int? clusterID;
+
+  const FaceWidget(
+    this.file,
+    this.face, {
+    this.person,
+    this.clusterID,
+    Key? key,
+  }) : super(key: key);
+
+  @override
+  Widget build(BuildContext context) {
+    return FutureBuilder<Uint8List?>(
+      future: getFaceCrop(),
+      builder: (context, snapshot) {
+        if (snapshot.hasData) {
+          final ImageProvider imageProvider = MemoryImage(snapshot.data!);
+          return GestureDetector(
+            onTap: () async {
+              log(
+                "FaceWidget is tapped, with person $person and clusterID $clusterID",
+                name: "FaceWidget",
+              );
+              if (person == null && clusterID == null) {
+                return;
+              }
+              if (person != null) {
+                await Navigator.of(context).push(
+                  MaterialPageRoute(
+                    builder: (context) => PeoplePage(
+                      person: person!,
+                    ),
+                  ),
+                );
+              } else if (clusterID != null) {
+                final fileIdsToClusterIds =
+                    await FaceMLDataDB.instance.getFileIdToClusterIds();
+                final files = await SearchService.instance.getAllFiles();
+                final clusterFiles = files
+                    .where(
+                      (file) =>
+                          fileIdsToClusterIds[file.uploadedFileID]
+                              ?.contains(clusterID) ??
+                          false,
+                    )
+                    .toList();
+                await Navigator.of(context).push(
+                  MaterialPageRoute(
+                    builder: (context) => ClusterPage(
+                      clusterFiles,
+                      cluserID: clusterID!,
+                    ),
+                  ),
+                );
+              }
+            },
+            child: Column(
+              children: [
+                ClipOval(
+                  child: SizedBox(
+                    width: 60,
+                    height: 60,
+                    child: Image(
+                      image: imageProvider,
+                      fit: BoxFit.cover,
+                    ),
+                  ),
+                ),
+                const SizedBox(height: 8),
+                if (person != null)
+                  Text(
+                    person!.attr.name.trim(),
+                    style: Theme.of(context).textTheme.bodySmall,
+                    overflow: TextOverflow.ellipsis,
+                    maxLines: 1,
+                  ),
+              ],
+            ),
+          );
+        } else {
+          if (snapshot.connectionState == ConnectionState.waiting) {
+            return const ClipOval(
+              child: SizedBox(
+                width: 60, // Ensure consistent sizing
+                height: 60,
+                child: CircularProgressIndicator(),
+              ),
+            );
+          }
+          if (snapshot.hasError) {
+            log('Error getting face: ${snapshot.error}');
+          }
+          return const ClipOval(
+            child: SizedBox(
+              width: 60, // Ensure consistent sizing
+              height: 60,
+              child: NoThumbnailWidget(),
+            ),
+          );
+        }
+      },
+    );
+  }
+
+  Future<Uint8List?> getFaceCrop() async {
+    try {
+      final Uint8List? cachedFace = faceCropCache.get(face.faceID);
+      if (cachedFace != null) {
+        return cachedFace;
+      }
+      final faceCropCacheFile = cachedFaceCropPath(face.faceID);
+      if ((await faceCropCacheFile.exists())) {
+        final data = await faceCropCacheFile.readAsBytes();
+        faceCropCache.put(face.faceID, data);
+        return data;
+      }
+
+      final result = await pool.withResource(
+        () async => await getFaceCrops(
+          file,
+          {
+            face.faceID: face.detection.box,
+          },
+        ),
+      );
+      final Uint8List? computedCrop = result?[face.faceID];
+      if (computedCrop != null) {
+        faceCropCache.put(face.faceID, computedCrop);
+        faceCropCacheFile.writeAsBytes(computedCrop).ignore();
+      }
+      return computedCrop;
+    } catch (e, s) {
+      log(
+        "Error getting face for faceID: ${face.faceID}",
+        error: e,
+        stackTrace: s,
+      );
+      return null;
+    }
+  }
+}

mobile/lib/ui/viewer/file_details/faces_item_widget.dart

@@ -0,0 +1,79 @@
+import "package:flutter/material.dart";
+import "package:logging/logging.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/face.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/models/file/file.dart";
+import "package:photos/ui/components/buttons/chip_button_widget.dart";
+import "package:photos/ui/components/info_item_widget.dart";
+import "package:photos/ui/viewer/file_details/face_widget.dart";
+
+class FacesItemWidget extends StatelessWidget {
+  final EnteFile file;
+  const FacesItemWidget(this.file, {super.key});
+
+  @override
+  Widget build(BuildContext context) {
+    return InfoItemWidget(
+      key: const ValueKey("Faces"),
+      leadingIcon: Icons.face_retouching_natural_outlined,
+      subtitleSection: _faceWidgets(context, file),
+      hasChipButtons: true,
+    );
+  }
+
+  Future<List<Widget>> _faceWidgets(
+    BuildContext context,
+    EnteFile file,
+  ) async {
+    try {
+      if (file.uploadedFileID == null) {
+        return [
+          const ChipButtonWidget(
+            "File not uploaded yet",
+            noChips: true,
+          ),
+        ];
+      }
+
+      final List<Face> faces = await FaceMLDataDB.instance
+          .getFacesForGivenFileID(file.uploadedFileID!);
+      if (faces.isEmpty || faces.every((face) => face.score < 0.5)) {
+        return [
+          const ChipButtonWidget(
+            "No faces found",
+            noChips: true,
+          ),
+        ];
+      }
+
+      // Sort the faces by score in descending order, so that the highest scoring face is first.
+      faces.sort((Face a, Face b) => b.score.compareTo(a.score));
+
+      // TODO: add deduplication of faces of same person
+      final faceIdsToClusterIds = await FaceMLDataDB.instance
+          .getFaceIdsToClusterIds(faces.map((face) => face.faceID));
+      final (clusterIDToPerson, personIdToPerson) =
+          await FaceMLDataDB.instance.getClusterIdToPerson();
+
+      final faceWidgets = <FaceWidget>[];
+      for (final Face face in faces) {
+        final int? clusterID = faceIdsToClusterIds[face.faceID];
+        final Person? person = clusterIDToPerson[clusterID];
+        faceWidgets.add(
+          FaceWidget(
+            file,
+            face,
+            clusterID: clusterID,
+            person: person,
+          ),
+        );
+      }
+
+      return faceWidgets;
+    } catch (e, s) {
+      Logger("FacesItemWidget").info(e, s);
+      return <FaceWidget>[];
+    }
+  }
+}

mobile/lib/ui/viewer/file_details/objects_item_widget.dart

@@ -27,6 +27,7 @@ class ObjectsItemWidget extends StatelessWidget {
     try {
       final chipButtons = <ChipButtonWidget>[];
       var objectTags = <String, double>{};
+
       // final thumbnail = await getThumbnail(file);
       // if (thumbnail != null) {
       //   objectTags = await ObjectDetectionService.instance.predict(thumbnail);

mobile/lib/ui/viewer/people/add_person_action_sheet.dart

@@ -0,0 +1,301 @@
+import "dart:async";
+import "dart:developer";
+import "dart:math" as math;
+
+import 'package:flutter/material.dart';
+import "package:logging/logging.dart";
+import 'package:modal_bottom_sheet/modal_bottom_sheet.dart';
+import "package:photos/core/event_bus.dart";
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/generated/l10n.dart";
+import "package:photos/services/face_ml/feedback/cluster_feedback.dart";
+import 'package:photos/theme/colors.dart';
+import 'package:photos/theme/ente_theme.dart';
+import 'package:photos/ui/common/loading_widget.dart';
+import 'package:photos/ui/components/bottom_of_title_bar_widget.dart';
+import 'package:photos/ui/components/buttons/button_widget.dart';
+import 'package:photos/ui/components/models/button_type.dart';
+import "package:photos/ui/components/text_input_widget.dart";
+import 'package:photos/ui/components/title_bar_title_widget.dart';
+import "package:photos/ui/viewer/people/new_person_item_widget.dart";
+import "package:photos/ui/viewer/people/person_row_item.dart";
+import "package:photos/utils/dialog_util.dart";
+import "package:photos/utils/toast_util.dart";
+import "package:uuid/uuid.dart";
+
+enum PersonActionType {
+  assignPerson,
+}
+
+String _actionName(
+  BuildContext context,
+  PersonActionType type,
+) {
+  String text = "";
+  switch (type) {
+    case PersonActionType.assignPerson:
+      text = "Add name";
+      break;
+  }
+  return text;
+}
+
+Future<dynamic> showAssignPersonAction(
+  BuildContext context, {
+  required int clusterID,
+  PersonActionType actionType = PersonActionType.assignPerson,
+  bool showOptionToCreateNewAlbum = true,
+}) {
+  return showBarModalBottomSheet(
+    context: context,
+    builder: (context) {
+      return PersonActionSheet(
+        actionType: actionType,
+        showOptionToCreateNewAlbum: showOptionToCreateNewAlbum,
+        cluserID: clusterID,
+      );
+    },
+    shape: const RoundedRectangleBorder(
+      side: BorderSide(width: 0),
+      borderRadius: BorderRadius.vertical(
+        top: Radius.circular(5),
+      ),
+    ),
+    topControl: const SizedBox.shrink(),
+    backgroundColor: getEnteColorScheme(context).backgroundElevated,
+    barrierColor: backdropFaintDark,
+    enableDrag: false,
+  );
+}
+
+class PersonActionSheet extends StatefulWidget {
+  final PersonActionType actionType;
+  final int cluserID;
+  final bool showOptionToCreateNewAlbum;
+  const PersonActionSheet({
+    required this.actionType,
+    required this.cluserID,
+    required this.showOptionToCreateNewAlbum,
+    super.key,
+  });
+
+  @override
+  State<PersonActionSheet> createState() => _PersonActionSheetState();
+}
+
+class _PersonActionSheetState extends State<PersonActionSheet> {
+  static const int cancelButtonSize = 80;
+  String _searchQuery = "";
+
+  @override
+  void initState() {
+    super.initState();
+  }
+
+  @override
+  Widget build(BuildContext context) {
+    final bottomInset = MediaQuery.of(context).viewInsets.bottom;
+    final isKeyboardUp = bottomInset > 100;
+    return Padding(
+      padding: EdgeInsets.only(
+        bottom: isKeyboardUp ? bottomInset - cancelButtonSize : 0,
+      ),
+      child: Row(
+        mainAxisAlignment: MainAxisAlignment.center,
+        children: [
+          ConstrainedBox(
+            constraints: BoxConstraints(
+              maxWidth: math.min(428, MediaQuery.of(context).size.width),
+            ),
+            child: Padding(
+              padding: const EdgeInsets.fromLTRB(0, 32, 0, 8),
+              child: Column(
+                mainAxisSize: MainAxisSize.max,
+                children: [
+                  Expanded(
+                    child: Column(
+                      children: [
+                        BottomOfTitleBarWidget(
+                          title: TitleBarTitleWidget(
+                            title: _actionName(context, widget.actionType),
+                          ),
+                          // caption: 'Select or create a ',
+                        ),
+                        Padding(
+                          padding: const EdgeInsets.only(
+                            top: 16,
+                            left: 16,
+                            right: 16,
+                          ),
+                          child: TextInputWidget(
+                            hintText: 'Person name',
+                            prefixIcon: Icons.search_rounded,
+                            onChange: (value) {
+                              setState(() {
+                                _searchQuery = value;
+                              });
+                            },
+                            isClearable: true,
+                            shouldUnfocusOnClearOrSubmit: true,
+                            borderRadius: 2,
+                          ),
+                        ),
+                        _getPersonItems(),
+                      ],
+                    ),
+                  ),
+                  SafeArea(
+                    child: Container(
+                      //inner stroke of 1pt + 15 pts of top padding = 16 pts
+                      padding: const EdgeInsets.fromLTRB(16, 15, 16, 8),
+                      decoration: BoxDecoration(
+                        border: Border(
+                          top: BorderSide(
+                            color: getEnteColorScheme(context).strokeFaint,
+                          ),
+                        ),
+                      ),
+                      child: ButtonWidget(
+                        buttonType: ButtonType.secondary,
+                        buttonAction: ButtonAction.cancel,
+                        isInAlert: true,
+                        labelText: S.of(context).cancel,
+                      ),
+                    ),
+                  ),
+                ],
+              ),
+            ),
+          ),
+        ],
+      ),
+    );
+  }
+
+  Flexible _getPersonItems() {
+    return Flexible(
+      child: Padding(
+        padding: const EdgeInsets.fromLTRB(16, 24, 4, 0),
+        child: FutureBuilder<List<Person>>(
+          future: _getPersons(),
+          builder: (context, snapshot) {
+            if (snapshot.hasError) {
+              log("Error: ${snapshot.error} ${snapshot.stackTrace}}");
+              //Need to show an error on the UI here
+              return const SizedBox.shrink();
+            } else if (snapshot.hasData) {
+              final persons = snapshot.data as List<Person>;
+              final searchResults = _searchQuery.isNotEmpty
+                  ? persons
+                      .where(
+                        (element) => element.attr.name
+                            .toLowerCase()
+                            .contains(_searchQuery),
+                      )
+                      .toList()
+                  : persons;
+              final shouldShowCreateAlbum = widget.showOptionToCreateNewAlbum &&
+                  (_searchQuery.isEmpty || searchResults.isEmpty);
+
+              return Scrollbar(
+                thumbVisibility: true,
+                radius: const Radius.circular(2),
+                child: Padding(
+                  padding: const EdgeInsets.only(right: 12),
+                  child: ListView.separated(
+                    itemCount:
+                        searchResults.length + (shouldShowCreateAlbum ? 1 : 0),
+                    itemBuilder: (context, index) {
+                      if (index == 0 && shouldShowCreateAlbum) {
+                        return GestureDetector(
+                          child: const NewPersonItemWidget(),
+                          onTap: () async => {
+                            addNewPerson(
+                              context,
+                              initValue: _searchQuery.trim(),
+                              clusterID: widget.cluserID,
+                            ),
+                          },
+                        );
+                      }
+                      final person = searchResults[
+                          index - (shouldShowCreateAlbum ? 1 : 0)];
+                      return PersonRowItem(
+                        person: person,
+                        onTap: () async {
+                          await FaceMLDataDB.instance.assignClusterToPerson(
+                            personID: person.remoteID,
+                            clusterID: widget.cluserID,
+                          );
+                          Bus.instance.fire(PeopleChangedEvent());
+
+                          Navigator.pop(context, person);
+                        },
+                      );
+                    },
+                    separatorBuilder: (context, index) {
+                      return const SizedBox(height: 2);
+                    },
+                  ),
+                ),
+              );
+            } else {
+              return const EnteLoadingWidget();
+            }
+          },
+        ),
+      ),
+    );
+  }
+
+  Future<void> addNewPerson(
+    BuildContext context, {
+    String initValue = '',
+    required int clusterID,
+  }) async {
+    final result = await showTextInputDialog(
+      context,
+      title: "New person",
+      submitButtonLabel: 'Add',
+      hintText: 'Add name',
+      alwaysShowSuccessState: false,
+      initialValue: initValue,
+      textCapitalization: TextCapitalization.words,
+      onSubmit: (String text) async {
+        // indicates user cancelled the rename request
+        if (text.trim() == "") {
+          return;
+        }
+        try {
+          final String id = const Uuid().v4().toString();
+          final Person p = Person(
+            id,
+            PersonAttr(name: text, faces: <String>[]),
+          );
+          await FaceMLDataDB.instance.insert(p, clusterID);
+          final bool extraPhotosFound =
+              await ClusterFeedbackService.instance.checkAndDoAutomaticMerges(p);
+          if (extraPhotosFound) {
+            showShortToast(context, "Extra photos found for $text");
+          }
+          Bus.instance.fire(PeopleChangedEvent());
+          Navigator.pop(context, p);
+          log("inserted person");
+        } catch (e, s) {
+          Logger("_PersonActionSheetState")
+              .severe("Failed to rename album", e, s);
+          rethrow;
+        }
+      },
+    );
+    if (result is Exception) {
+      await showGenericErrorDialog(context: context, error: result);
+    }
+  }
+
+  Future<List<Person>> _getPersons() async {
+    return FaceMLDataDB.instance.getPeople();
+  }
+}

mobile/lib/ui/viewer/people/cluster_page.dart

@@ -0,0 +1,140 @@
+import "dart:async";
+
+import 'package:flutter/material.dart';
+import 'package:photos/core/event_bus.dart';
+import 'package:photos/events/files_updated_event.dart';
+import 'package:photos/events/local_photos_updated_event.dart';
+import "package:photos/face/model/person.dart";
+import 'package:photos/models/file/file.dart';
+import 'package:photos/models/file_load_result.dart';
+import 'package:photos/models/gallery_type.dart';
+import 'package:photos/models/selected_files.dart';
+import 'package:photos/ui/viewer/actions/file_selection_overlay_bar.dart';
+import 'package:photos/ui/viewer/gallery/gallery.dart';
+import 'package:photos/ui/viewer/gallery/gallery_app_bar_widget.dart';
+import "package:photos/ui/viewer/people/add_person_action_sheet.dart";
+import "package:photos/ui/viewer/people/people_page.dart";
+import "package:photos/ui/viewer/search/result/search_result_page.dart";
+import "package:photos/utils/navigation_util.dart";
+import "package:photos/utils/toast_util.dart";
+
+class ClusterPage extends StatefulWidget {
+  final List<EnteFile> searchResult;
+  final bool enableGrouping;
+  final String tagPrefix;
+  final int cluserID;
+  final Person? personID;
+
+  static const GalleryType appBarType = GalleryType.cluster;
+  static const GalleryType overlayType = GalleryType.cluster;
+
+  const ClusterPage(
+    this.searchResult, {
+    this.enableGrouping = true,
+    this.tagPrefix = "",
+    required this.cluserID,
+    this.personID,
+    Key? key,
+  }) : super(key: key);
+
+  @override
+  State<ClusterPage> createState() => _ClusterPageState();
+}
+
+class _ClusterPageState extends State<ClusterPage> {
+  final _selectedFiles = SelectedFiles();
+  late final List<EnteFile> files;
+  late final StreamSubscription<LocalPhotosUpdatedEvent> _filesUpdatedEvent;
+
+  @override
+  void initState() {
+    super.initState();
+    files = widget.searchResult;
+    _filesUpdatedEvent =
+        Bus.instance.on<LocalPhotosUpdatedEvent>().listen((event) {
+      if (event.type == EventType.deletedFromDevice ||
+          event.type == EventType.deletedFromEverywhere ||
+          event.type == EventType.deletedFromRemote ||
+          event.type == EventType.hide) {
+        for (var updatedFile in event.updatedFiles) {
+          files.remove(updatedFile);
+        }
+        setState(() {});
+      }
+    });
+  }
+
+  @override
+  void dispose() {
+    _filesUpdatedEvent.cancel();
+    super.dispose();
+  }
+
+  @override
+  Widget build(BuildContext context) {
+    final gallery = Gallery(
+      asyncLoader: (creationStartTime, creationEndTime, {limit, asc}) {
+        final result = files
+            .where(
+              (file) =>
+                  file.creationTime! >= creationStartTime &&
+                  file.creationTime! <= creationEndTime,
+            )
+            .toList();
+        return Future.value(
+          FileLoadResult(
+            result,
+            result.length < files.length,
+          ),
+        );
+      },
+      reloadEvent: Bus.instance.on<LocalPhotosUpdatedEvent>(),
+      removalEventTypes: const {
+        EventType.deletedFromRemote,
+        EventType.deletedFromEverywhere,
+        EventType.hide,
+      },
+      tagPrefix: widget.tagPrefix + widget.tagPrefix,
+      selectedFiles: _selectedFiles,
+      enableFileGrouping: widget.enableGrouping,
+      initialFiles: [widget.searchResult.first],
+    );
+    return Scaffold(
+      appBar: PreferredSize(
+        preferredSize: const Size.fromHeight(50.0),
+        child: GestureDetector(
+          onTap: () async {
+            if (widget.personID == null) {
+              final result = await showAssignPersonAction(
+                context,
+                clusterID: widget.cluserID,
+              );
+              if (result != null && result is Person) {
+                Navigator.pop(context);
+                // ignore: unawaited_futures
+                routeToPage(context, PeoplePage(person: result));
+              }
+            } else {
+              showShortToast(context, "11No personID or clusterID");
+            }
+          },
+          child: GalleryAppBarWidget(
+            SearchResultPage.appBarType,
+            widget.personID != null ? widget.personID!.attr.name : "Add name",
+            _selectedFiles,
+          ),
+        ),
+      ),
+      body: Stack(
+        alignment: Alignment.bottomCenter,
+        children: [
+          gallery,
+          FileSelectionOverlayBar(
+            ClusterPage.overlayType,
+            _selectedFiles,
+          ),
+        ],
+      ),
+    );
+  }
+}

mobile/lib/ui/viewer/people/new_person_item_widget.dart

@@ -0,0 +1,73 @@
+import 'package:dotted_border/dotted_border.dart';
+import 'package:flutter/material.dart';
+import 'package:photos/theme/ente_theme.dart';
+
+///https://www.figma.com/file/SYtMyLBs5SAOkTbfMMzhqt/ente-Visual-Design?node-id=10854%3A57947&t=H5AvR79OYDnB9ekw-4
+class NewPersonItemWidget extends StatelessWidget {
+  const NewPersonItemWidget({
+    super.key,
+  });
+
+  @override
+  Widget build(BuildContext context) {
+    final textTheme = getEnteTextTheme(context);
+    final colorScheme = getEnteColorScheme(context);
+    const sideOfThumbnail = 60.0;
+    return LayoutBuilder(
+      builder: (context, constraints) {
+        return Stack(
+          alignment: Alignment.center,
+          children: [
+            Row(
+              children: [
+                ClipRRect(
+                  borderRadius: const BorderRadius.horizontal(
+                    left: Radius.circular(4),
+                  ),
+                  child: SizedBox(
+                    height: sideOfThumbnail,
+                    width: sideOfThumbnail,
+                    child: Icon(
+                      Icons.add_outlined,
+                      color: colorScheme.strokeMuted,
+                    ),
+                  ),
+                ),
+                Padding(
+                  padding: const EdgeInsets.only(left: 12),
+                  child: Text(
+                    'Add person',
+                    style:
+                        textTheme.body.copyWith(color: colorScheme.textMuted),
+                  ),
+                ),
+              ],
+            ),
+            IgnorePointer(
+              child: DottedBorder(
+                dashPattern: const [4],
+                color: colorScheme.strokeFainter,
+                strokeWidth: 1,
+                padding: const EdgeInsets.all(0),
+                borderType: BorderType.RRect,
+                radius: const Radius.circular(4),
+                child: SizedBox(
+                  //Have to decrease the height and width by 1 pt as the stroke
+                  //dotted border gives is of strokeAlign.center, so 0.5 inside and
+                  // outside. Here for the row, stroke should be inside so we
+                  //decrease the size of this sizedBox by 1 (so it shrinks 0.5 from
+                  //every side) so that the strokeAlign.center of this sizedBox
+                  //looks like a strokeAlign.inside in the row.
+                  height: sideOfThumbnail - 1,
+                  //This width will work for this only if the row widget takes up the
+                  //full size it's parent (stack).
+                  width: constraints.maxWidth - 1,
+                ),
+              ),
+            ),
+          ],
+        );
+      },
+    );
+  }
+}

mobile/lib/ui/viewer/people/people_app_bar.dart

@@ -0,0 +1,256 @@
+import 'dart:async';
+
+import "package:flutter/cupertino.dart";
+import 'package:flutter/material.dart';
+import 'package:logging/logging.dart';
+import 'package:photos/core/configuration.dart';
+import 'package:photos/core/event_bus.dart';
+import "package:photos/events/people_changed_event.dart";
+import 'package:photos/events/subscription_purchased_event.dart';
+import "package:photos/face/db.dart";
+import "package:photos/face/model/person.dart";
+import "package:photos/generated/l10n.dart";
+import 'package:photos/models/gallery_type.dart';
+import 'package:photos/models/selected_files.dart';
+import 'package:photos/services/collections_service.dart';
+import 'package:photos/ui/actions/collection/collection_sharing_actions.dart';
+import "package:photos/ui/viewer/people/person_cluserts.dart";
+import "package:photos/ui/viewer/people/person_cluster_suggestion.dart";
+import "package:photos/utils/dialog_util.dart";
+
+class PeopleAppBar extends StatefulWidget {
+  final GalleryType type;
+  final String? title;
+  final SelectedFiles selectedFiles;
+  final Person person;
+
+  const PeopleAppBar(
+    this.type,
+    this.title,
+    this.selectedFiles,
+    this.person, {
+    Key? key,
+  }) : super(key: key);
+
+  @override
+  State<PeopleAppBar> createState() => _AppBarWidgetState();
+}
+
+enum PeoplPopupAction {
+  rename,
+  setCover,
+  viewPhotos,
+  confirmPhotos,
+  hide,
+}
+
+class _AppBarWidgetState extends State<PeopleAppBar> {
+  final _logger = Logger("_AppBarWidgetState");
+  late StreamSubscription _userAuthEventSubscription;
+  late Function() _selectedFilesListener;
+  String? _appBarTitle;
+  late CollectionActions collectionActions;
+  final GlobalKey shareButtonKey = GlobalKey();
+  bool isQuickLink = false;
+  late GalleryType galleryType;
+
+  @override
+  void initState() {
+    super.initState();
+    _selectedFilesListener = () {
+      setState(() {});
+    };
+    collectionActions = CollectionActions(CollectionsService.instance);
+    widget.selectedFiles.addListener(_selectedFilesListener);
+    _userAuthEventSubscription =
+        Bus.instance.on<SubscriptionPurchasedEvent>().listen((event) {
+      setState(() {});
+    });
+    _appBarTitle = widget.title;
+    galleryType = widget.type;
+  }
+
+  @override
+  void dispose() {
+    _userAuthEventSubscription.cancel();
+    widget.selectedFiles.removeListener(_selectedFilesListener);
+    super.dispose();
+  }
+
+  @override
+  Widget build(BuildContext context) {
+    return AppBar(
+      elevation: 0,
+      centerTitle: false,
+      title: Text(
+        _appBarTitle!,
+        style:
+            Theme.of(context).textTheme.headlineSmall!.copyWith(fontSize: 16),
+        maxLines: 2,
+        overflow: TextOverflow.ellipsis,
+      ),
+      actions: _getDefaultActions(context),
+    );
+  }
+
+  Future<dynamic> _renameAlbum(BuildContext context) async {
+    final result = await showTextInputDialog(
+      context,
+      title: 'Rename',
+      submitButtonLabel: S.of(context).done,
+      hintText: S.of(context).enterAlbumName,
+      alwaysShowSuccessState: true,
+      initialValue: widget.person.attr.name,
+      textCapitalization: TextCapitalization.words,
+      onSubmit: (String text) async {
+        // indicates user cancelled the rename request
+        if (text == "" || text == _appBarTitle!) {
+          return;
+        }
+
+        try {
+          final updatePerson = widget.person
+              .copyWith(attr: widget.person.attr.copyWith(name: text));
+          await FaceMLDataDB.instance.updatePerson(updatePerson);
+          if (mounted) {
+            _appBarTitle = text;
+            setState(() {});
+          }
+          Bus.instance.fire(PeopleChangedEvent());
+        } catch (e, s) {
+          _logger.severe("Failed to rename album", e, s);
+          rethrow;
+        }
+      },
+    );
+    if (result is Exception) {
+      await showGenericErrorDialog(context: context, error: result);
+    }
+  }
+
+  List<Widget> _getDefaultActions(BuildContext context) {
+    final List<Widget> actions = <Widget>[];
+    // If the user has selected files, don't show any actions
+    if (widget.selectedFiles.files.isNotEmpty ||
+        !Configuration.instance.hasConfiguredAccount()) {
+      return actions;
+    }
+
+    final List<PopupMenuItem<PeoplPopupAction>> items = [];
+
+    items.addAll(
+      [
+        PopupMenuItem(
+          value: PeoplPopupAction.rename,
+          child: Row(
+            children: [
+              const Icon(Icons.edit),
+              const Padding(
+                padding: EdgeInsets.all(8),
+              ),
+              Text(S.of(context).rename),
+            ],
+          ),
+        ),
+        // PopupMenuItem(
+        //   value: PeoplPopupAction.setCover,
+        //   child: Row(
+        //     children: [
+        //       const Icon(Icons.image_outlined),
+        //       const Padding(
+        //         padding: EdgeInsets.all(8),
+        //       ),
+        //       Text(S.of(context).setCover),
+        //     ],
+        //   ),
+        // ),
+        // PopupMenuItem(
+        //   value: PeoplPopupAction.rename,
+        //   child: Row(
+        //     children: [
+        //       const Icon(Icons.visibility_off),
+        //       const Padding(
+        //         padding: EdgeInsets.all(8),
+        //       ),
+        //       Text(S.of(context).hide),
+        //     ],
+        //   ),
+        // ),
+        const PopupMenuItem(
+          value: PeoplPopupAction.viewPhotos,
+          child: Row(
+            children: [
+              Icon(Icons.view_array_outlined),
+              Padding(
+                padding: EdgeInsets.all(8),
+              ),
+              Text('View confirmed photos'),
+            ],
+          ),
+        ),
+        const PopupMenuItem(
+          value: PeoplPopupAction.confirmPhotos,
+          child: Row(
+            children: [
+              Icon(CupertinoIcons.square_stack_3d_down_right),
+              Padding(
+                padding: EdgeInsets.all(8),
+              ),
+              Text('Review suggestions'),
+            ],
+          ),
+        ),
+      ],
+    );
+
+    if (items.isNotEmpty) {
+      actions.add(
+        PopupMenuButton(
+          itemBuilder: (context) {
+            return items;
+          },
+          onSelected: (PeoplPopupAction value) async {
+            if (value == PeoplPopupAction.viewPhotos) {
+              // ignore: unawaited_futures
+              unawaited(
+                Navigator.of(context).push(
+                  MaterialPageRoute(
+                    builder: (context) => PersonClusters(widget.person),
+                  ),
+                ),
+              );
+            } else if (value == PeoplPopupAction.confirmPhotos) {
+              // ignore: unawaited_futures
+              unawaited(
+                Navigator.of(context).push(
+                  MaterialPageRoute(
+                    builder: (context) =>
+                        PersonReviewClusterSuggestion(widget.person),
+                  ),
+                ),
+              );
+            } else if (value == PeoplPopupAction.rename) {
+              await _renameAlbum(context);
+            } else if (value == PeoplPopupAction.setCover) {
+              await setCoverPhoto(context);
+            } else if (value == PeoplPopupAction.hide) {
+              // ignore: unawaited_futures
+            }
+          },
+        ),
+      );
+    }
+
+    return actions;
+  }
+
+  Future<void> setCoverPhoto(BuildContext context) async {
+    // final int? coverPhotoID = await showPickCoverPhotoSheet(
+    //   context,
+    //   widget.collection!,
+    // );
+    // if (coverPhotoID != null) {
+    //   unawaited(changeCoverPhoto(context, widget.collection!, coverPhotoID));
+    // }
+  }
+}

mobile/lib/ui/viewer/people/people_page.dart

@@ -0,0 +1,155 @@
+import "dart:async";
+import "dart:developer";
+
+import 'package:flutter/material.dart';
+import "package:logging/logging.dart";
+import 'package:photos/core/event_bus.dart';
+import 'package:photos/events/files_updated_event.dart';
+import 'package:photos/events/local_photos_updated_event.dart';
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/face/model/person.dart";
+import 'package:photos/models/file/file.dart';
+import 'package:photos/models/file_load_result.dart';
+import 'package:photos/models/gallery_type.dart';
+import 'package:photos/models/selected_files.dart';
+import "package:photos/services/search_service.dart";
+import 'package:photos/ui/viewer/actions/file_selection_overlay_bar.dart';
+import 'package:photos/ui/viewer/gallery/gallery.dart';
+import "package:photos/ui/viewer/people/people_app_bar.dart";
+
+class PeoplePage extends StatefulWidget {
+  final String tagPrefix;
+  final Person person;
+
+  static const GalleryType appBarType = GalleryType.peopleTag;
+  static const GalleryType overlayType = GalleryType.peopleTag;
+
+  const PeoplePage({
+    this.tagPrefix = "",
+    required this.person,
+    Key? key,
+  }) : super(key: key);
+
+  @override
+  State<PeoplePage> createState() => _PeoplePageState();
+}
+
+class _PeoplePageState extends State<PeoplePage> {
+  final Logger _logger = Logger("_PeoplePageState");
+  final _selectedFiles = SelectedFiles();
+  List<EnteFile>? files;
+
+  late final StreamSubscription<LocalPhotosUpdatedEvent> _filesUpdatedEvent;
+  late final StreamSubscription<PeopleChangedEvent> _peopleChangedEvent;
+
+  @override
+  void initState() {
+    super.initState();
+    _peopleChangedEvent = Bus.instance.on<PeopleChangedEvent>().listen((event) {
+      setState(() {});
+    });
+
+    _filesUpdatedEvent =
+        Bus.instance.on<LocalPhotosUpdatedEvent>().listen((event) {
+      if (event.type == EventType.deletedFromDevice ||
+          event.type == EventType.deletedFromEverywhere ||
+          event.type == EventType.deletedFromRemote ||
+          event.type == EventType.hide) {
+        for (var updatedFile in event.updatedFiles) {
+          files?.remove(updatedFile);
+        }
+        setState(() {});
+      }
+    });
+  }
+
+  Future<List<EnteFile>> loadPersonFiles() async {
+    log("loadPersonFiles");
+    final result = await SearchService.instance
+        .getClusterFilesForPersonID(widget.person.remoteID);
+    final List<EnteFile> resultFiles = [];
+    for (final e in result.entries) {
+      resultFiles.addAll(e.value);
+    }
+    files = resultFiles;
+    return resultFiles;
+  }
+
+  @override
+  void dispose() {
+    _filesUpdatedEvent.cancel();
+    _peopleChangedEvent.cancel();
+    super.dispose();
+  }
+
+  @override
+  Widget build(BuildContext context) {
+    _logger.info("Building for ${widget.person.attr.name}");
+    return Scaffold(
+      appBar: PreferredSize(
+        preferredSize: const Size.fromHeight(50.0),
+        child: PeopleAppBar(
+          GalleryType.peopleTag,
+          widget.person.attr.name,
+          _selectedFiles,
+          widget.person,
+        ),
+      ),
+      body: Stack(
+        alignment: Alignment.bottomCenter,
+        children: [
+          FutureBuilder<List<EnteFile>>(
+            future: loadPersonFiles(),
+            builder: (context, snapshot) {
+              if (snapshot.hasData) {
+                final personFiles = snapshot.data as List<EnteFile>;
+                return Gallery(
+                  asyncLoader: (
+                    creationStartTime,
+                    creationEndTime, {
+                    limit,
+                    asc,
+                  }) async {
+                    final result = await loadPersonFiles();
+                    return Future.value(
+                      FileLoadResult(
+                        result,
+                        false,
+                      ),
+                    );
+                  },
+                  reloadEvent: Bus.instance.on<LocalPhotosUpdatedEvent>(),
+                  forceReloadEvents: [
+                    Bus.instance.on<PeopleChangedEvent>(),
+                  ],
+                  removalEventTypes: const {
+                    EventType.deletedFromRemote,
+                    EventType.deletedFromEverywhere,
+                    EventType.hide,
+                  },
+                  tagPrefix: widget.tagPrefix + widget.tagPrefix,
+                  selectedFiles: _selectedFiles,
+                  initialFiles:
+                      personFiles.isNotEmpty ? [personFiles.first] : [],
+                );
+              } else if (snapshot.hasError) {
+                log("Error: ${snapshot.error} ${snapshot.stackTrace}}");
+                //Need to show an error on the UI here
+                return const SizedBox.shrink();
+              } else {
+                return const Center(
+                  child: CircularProgressIndicator(),
+                );
+              }
+            },
+          ),
+          FileSelectionOverlayBar(
+            PeoplePage.overlayType,
+            _selectedFiles,
+            person: widget.person,
+          ),
+        ],
+      ),
+    );
+  }
+}