[mob][photos] Image embedding caching perf (#5459)

## Description Made changes to more efficiently handle image embeddings caching, keeping them cached only in MLComputer and only for limited times when the user needs it. ## Tests Tested in debug mode on my pixel phone.
2025-03-28 14:13:42 +05:30
parent 60781ba86b d11ff14ecd
commit 360223bd2f
10 changed files with 269 additions and 154 deletions
--- a/mobile/lib/db/ml/db.dart
+++ b/mobile/lib/db/ml/db.dart
@@ -1166,8 +1166,21 @@ class MLDataDB with SqlDbBase implements IMLDataDB<int> {
  Future<List<EmbeddingVector>> getAllClipVectors() async {
    Logger("ClipDB").info("reading all embeddings from DB");
    final db = await instance.asyncDB;
-    final results = await db.getAll('SELECT * FROM $clipTable');
-    return _convertToVectors(results);
+    final results = await db
+        .getAll('SELECT $fileIDColumn, $embeddingColumn FROM $clipTable');
+
+    // Convert rows to vectors
+    final List<EmbeddingVector> embeddings = [];
+    for (final result in results) {
+      // Convert to EmbeddingVector
+      final embedding = EmbeddingVector(
+        fileID: result[fileIDColumn],
+        embedding: Float32List.view(result[embeddingColumn].buffer),
+      );
+      if (embedding.isEmpty) continue;
+      embeddings.add(embedding);
+    }
+    return embeddings;
  }

  // Get indexed FileIDs
@@ -1229,23 +1242,6 @@ class MLDataDB with SqlDbBase implements IMLDataDB<int> {
    Bus.instance.fire(EmbeddingUpdatedEvent());
  }

-  List<EmbeddingVector> _convertToVectors(List<Map<String, dynamic>> results) {
-    final List<EmbeddingVector> embeddings = [];
-    for (final result in results) {
-      final embedding = _getVectorFromRow(result);
-      if (embedding.isEmpty) continue;
-      embeddings.add(embedding);
-    }
-    return embeddings;
-  }
-
-  EmbeddingVector _getVectorFromRow(Map<String, dynamic> row) {
-    final fileID = row[fileIDColumn] as int;
-    final bytes = row[embeddingColumn] as Uint8List;
-    final list = Float32List.view(bytes.buffer);
-    return EmbeddingVector(fileID: fileID, embedding: list);
-  }
-
  List<Object?> _getRowFromEmbedding(ClipEmbedding embedding) {
    return [
      embedding.fileID,
--- a/mobile/lib/services/isolate_functions.dart
+++ b/mobile/lib/services/isolate_functions.dart
@@ -5,13 +5,17 @@ import "package:ml_linalg/linalg.dart";
 import "package:photos/models/ml/face/box.dart";
 import "package:photos/models/ml/vector.dart";
 import "package:photos/services/machine_learning/face_ml/face_clustering/face_clustering_service.dart";
+import "package:photos/services/machine_learning/ml_constants.dart";
 import "package:photos/services/machine_learning/ml_model.dart";
 import "package:photos/services/machine_learning/ml_result.dart";
 import "package:photos/services/machine_learning/semantic_search/clip/clip_text_encoder.dart";
 import "package:photos/services/machine_learning/semantic_search/clip/clip_text_tokenizer.dart";
+import "package:photos/services/machine_learning/semantic_search/query_result.dart";
 import "package:photos/utils/image_ml_util.dart";
 import "package:photos/utils/ml_util.dart";

+final Map<String, dynamic> _isolateCache = {};
+
 enum IsolateOperation {
  /// [MLIndexingIsolate]
  analyzeImage,
@@ -35,10 +39,15 @@ enum IsolateOperation {
  runClipText,

  /// [MLComputer]
-  compareEmbeddings,
+  computeBulkSimilarities,

  /// [FaceClusteringService]
  linearIncrementalClustering,
+
+  /// Cache operations
+  setIsolateCache,
+  clearIsolateCache,
+  clearAllIsolateCache,
 }

 /// WARNING: Only return primitives unless you know the method is only going
@@ -121,19 +130,31 @@ Future<dynamic> isolateFunction(
      return List<double>.from(textEmbedding, growable: false);

    /// MLComputer
-    case IsolateOperation.compareEmbeddings:
-      final List<EmbeddingVector> embeddings =
-          (args['embeddings'] as List<String>)
-              .map((jsonString) => EmbeddingVector.fromJsonString(jsonString))
-              .toList();
-      final otherEmbedding =
-          Vector.fromList(args['otherEmbedding'] as List<double>);
-      final Map<int, double> result = {};
-      for (final embedding in embeddings) {
-        final double similarity = embedding.vector.dot(otherEmbedding);
-        result[embedding.fileID] = similarity;
+    case IsolateOperation.computeBulkSimilarities:
+      final imageEmbeddings =
+          _isolateCache[imageEmbeddingsKey] as List<EmbeddingVector>;
+      final textEmbedding =
+          args["textQueryToEmbeddingMap"] as Map<String, List<double>>;
+      final minimumSimilarityMap =
+          args["minimumSimilarityMap"] as Map<String, double>;
+      final result = <String, List<QueryResult>>{};
+      for (final MapEntry<String, List<double>> entry
+          in textEmbedding.entries) {
+        final query = entry.key;
+        final textVector = Vector.fromList(entry.value);
+        final minimumSimilarity = minimumSimilarityMap[query]!;
+        final queryResults = <QueryResult>[];
+        for (final imageEmbedding in imageEmbeddings) {
+          final similarity = imageEmbedding.vector.dot(textVector);
+          if (similarity >= minimumSimilarity) {
+            queryResults.add(QueryResult(imageEmbedding.fileID, similarity));
+          }
+        }
+        queryResults
+            .sort((first, second) => second.score.compareTo(first.score));
+        result[query] = queryResults;
      }
-      return Map<int, double>.from(result);
+      return result;

    /// Cases for MLComputer end here

@@ -145,5 +166,27 @@ Future<dynamic> isolateFunction(
      return result;

    /// Cases for FaceClusteringService end here
+
+    /// Cases for Caching start here
+
+    /// Caching
+    case IsolateOperation.setIsolateCache:
+      final key = args['key'] as String;
+      final value = args['value'];
+      _isolateCache[key] = value;
+      return true;
+
+    /// Caching
+    case IsolateOperation.clearIsolateCache:
+      final key = args['key'] as String;
+      _isolateCache.remove(key);
+      return true;
+
+    /// Caching
+    case IsolateOperation.clearAllIsolateCache:
+      _isolateCache.clear();
+      return true;
+
+    /// Cases for Caching stop here
  }
 }
--- a/mobile/lib/services/isolate_service.dart
+++ b/mobile/lib/services/isolate_service.dart
@@ -3,6 +3,7 @@ import 'dart:isolate';

 import "package:dart_ui_isolate/dart_ui_isolate.dart";
 import "package:flutter/foundation.dart" show kDebugMode;
+import "package:flutter/services.dart";
 import "package:logging/logging.dart";
 import "package:photos/core/error-reporting/isolate_logging.dart";
 import "package:photos/models/base/id.dart";
@@ -13,7 +14,7 @@ abstract class SuperIsolate {
  Logger get logger;

  Timer? _inactivityTimer;
-  final Duration _inactivityDuration = const Duration(seconds: 120);
+  final Duration _inactivityDuration = const Duration(seconds: 60);
  int _activeTasks = 0;

  final _initIsolateLock = Lock();
@@ -36,15 +37,22 @@ abstract class SuperIsolate {

      _receivePort = ReceivePort();

+      // Get the root token before spawning the isolate
+      final rootToken = RootIsolateToken.instance;
+      if (rootToken == null && !isDartUiIsolate) {
+        logger.severe('Failed to get RootIsolateToken');
+        return;
+      }
+
      try {
        _isolate = isDartUiIsolate
            ? await DartUiIsolate.spawn(
                _isolateMain,
-                _receivePort.sendPort,
+                [_receivePort.sendPort, null],
              )
            : await Isolate.spawn(
                _isolateMain,
-                _receivePort.sendPort,
+                [_receivePort.sendPort, rootToken],
                debugName: isolateName,
              );
        _mainSendPort = await _receivePort.first as SendPort;
@@ -60,12 +68,18 @@ abstract class SuperIsolate {
  }

  @pragma('vm:entry-point')
-  static void _isolateMain(SendPort mainSendPort) async {
+  static void _isolateMain(List<dynamic> args) async {
+    final SendPort mainSendPort = args[0] as SendPort;
+    final RootIsolateToken? rootToken = args[1] as RootIsolateToken?;
+
    Logger.root.level = kDebugMode ? Level.ALL : Level.INFO;
    final IsolateLogger isolateLogger = IsolateLogger();
    Logger.root.onRecord.listen(isolateLogger.onLogRecordInIsolate);
    final receivePort = ReceivePort();
    mainSendPort.send(receivePort.sendPort);
+    if (rootToken != null) {
+      BackgroundIsolateBinaryMessenger.ensureInitialized(rootToken);
+    }

    receivePort.listen((message) async {
      final taskID = message[0] as String;
@@ -137,6 +151,24 @@ abstract class SuperIsolate {

  bool postFunctionlockStop(IsolateOperation operation) => false;

+  Future<void> cacheData(String key, dynamic value) async {
+    await runInIsolate(IsolateOperation.setIsolateCache, {
+      'key': key,
+      'value': value,
+    });
+  }
+
+  /// Clears specific data from the isolate's cache
+  Future<void> clearCachedData(String key) async {
+    await runInIsolate(IsolateOperation.clearIsolateCache, {
+      'key': key,
+    });
+  }
+
+  Future<void> clearAllCachedData() async {
+    await runInIsolate(IsolateOperation.clearAllIsolateCache, {});
+  }
+
  /// 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()`)
@@ -161,6 +193,7 @@ abstract class SuperIsolate {
  void _disposeIsolate() async {
    if (!_isIsolateSpawned) return;
    logger.info('Disposing isolate');
+    await clearAllCachedData();
    await onDispose();
    _isIsolateSpawned = false;
    _isolate.kill();
--- a/mobile/lib/services/machine_learning/face_thumbnail_generator.dart
+++ b/mobile/lib/services/machine_learning/face_thumbnail_generator.dart
@@ -0,0 +1,47 @@
+import 'dart:async';
+import 'dart:typed_data' show Uint8List;
+
+import "package:logging/logging.dart";
+import "package:photos/models/ml/face/box.dart";
+import "package:photos/services/isolate_functions.dart";
+import "package:photos/services/isolate_service.dart";
+import "package:photos/utils/image_ml_util.dart";
+
+class FaceThumbnailGenerator extends SuperIsolate {
+  @override
+  Logger get logger => _logger;
+  final _logger = Logger('FaceThumbnailGenerator');
+
+  @override
+  bool get isDartUiIsolate => true;
+
+  @override
+  String get isolateName => "FaceThumbnailGenerator";
+
+  @override
+  bool get shouldAutomaticDispose => true;
+
+  // Singleton pattern
+  FaceThumbnailGenerator._privateConstructor();
+  static final FaceThumbnailGenerator instance =
+      FaceThumbnailGenerator._privateConstructor();
+  factory FaceThumbnailGenerator() => instance;
+
+  /// Generates face thumbnails for all [faceBoxes] in [imageData].
+  ///
+  /// Uses [generateFaceThumbnailsUsingCanvas] inside the isolate.
+  Future<List<Uint8List>> generateFaceThumbnails(
+    String imagePath,
+    List<FaceBox> faceBoxes,
+  ) async {
+    final List<Map<String, dynamic>> faceBoxesJson =
+        faceBoxes.map((box) => box.toJson()).toList();
+    return await runInIsolate(
+      IsolateOperation.generateFaceThumbnails,
+      {
+        'imagePath': imagePath,
+        'faceBoxesList': faceBoxesJson,
+      },
+    ).then((value) => value.cast<Uint8List>());
+  }
+}
--- a/mobile/lib/services/machine_learning/ml_computer.dart
+++ b/mobile/lib/services/machine_learning/ml_computer.dart
@@ -1,15 +1,13 @@
 import 'dart:async';
-import 'dart:typed_data' show Uint8List;

 import "package:logging/logging.dart";
-import "package:ml_linalg/linalg.dart";
-import "package:photos/models/ml/face/box.dart";
 import "package:photos/models/ml/vector.dart";
 import "package:photos/services/isolate_functions.dart";
 import "package:photos/services/isolate_service.dart";
+import "package:photos/services/machine_learning/ml_constants.dart";
 import "package:photos/services/machine_learning/semantic_search/clip/clip_text_encoder.dart";
+import "package:photos/services/machine_learning/semantic_search/query_result.dart";
 import "package:photos/services/remote_assets_service.dart";
-import "package:photos/utils/image_ml_util.dart";
 import "package:synchronized/synchronized.dart";

 class MLComputer extends SuperIsolate {
@@ -20,7 +18,7 @@ class MLComputer extends SuperIsolate {
  final _initModelLock = Lock();

  @override
-  bool get isDartUiIsolate => true;
+  bool get isDartUiIsolate => false;

  @override
  String get isolateName => "MLComputerIsolate";
@@ -33,24 +31,6 @@ class MLComputer extends SuperIsolate {
  static final MLComputer instance = MLComputer._privateConstructor();
  factory MLComputer() => instance;

-  /// Generates face thumbnails for all [faceBoxes] in [imageData].
-  ///
-  /// Uses [generateFaceThumbnailsUsingCanvas] inside the isolate.
-  Future<List<Uint8List>> generateFaceThumbnails(
-    String imagePath,
-    List<FaceBox> faceBoxes,
-  ) async {
-    final List<Map<String, dynamic>> faceBoxesJson =
-        faceBoxes.map((box) => box.toJson()).toList();
-    return await runInIsolate(
-      IsolateOperation.generateFaceThumbnails,
-      {
-        'imagePath': imagePath,
-        'faceBoxesList': faceBoxesJson,
-      },
-    ).then((value) => value.cast<Uint8List>());
-  }
-
  Future<List<double>> runClipText(String query) async {
    try {
      await _ensureLoadedClipTextModel();
@@ -66,19 +46,6 @@ class MLComputer extends SuperIsolate {
    }
  }

-  Future<Map<int, double>> compareEmbeddings(List<EmbeddingVector> embeddings, Vector otherEmbedding) async {
-    try {
-      final fileIdToSimilarity = await runInIsolate(IsolateOperation.compareEmbeddings, {
-        "embeddings": embeddings.map((e) => e.toJsonString()).toList(),
-        "otherEmbedding": otherEmbedding.toList(),
-      }) as Map<int, double>;
-      return fileIdToSimilarity;
-    } catch (e, s) {
-      _logger.severe("Could not compare embeddings MLComputer isolate", e, s);
-      rethrow;
-    }
-  }
-
  Future<void> _ensureLoadedClipTextModel() async {
    return _initModelLock.synchronized(() async {
      if (ClipTextEncoder.instance.isInitialized) return;
@@ -114,4 +81,61 @@ class MLComputer extends SuperIsolate {
      }
    });
  }
+
+  Future<Map<String, List<QueryResult>>> computeBulkSimilarities(
+    Map<String, List<double>> textQueryToEmbeddingMap,
+    Map<String, double> minimumSimilarityMap,
+  ) async {
+    try {
+      final queryToResults =
+          await runInIsolate(IsolateOperation.computeBulkSimilarities, {
+        "textQueryToEmbeddingMap": textQueryToEmbeddingMap,
+        "minimumSimilarityMap": minimumSimilarityMap,
+      }) as Map<String, List<QueryResult>>;
+      return queryToResults;
+    } catch (e, s) {
+      _logger.severe(
+        "Could not bulk compare embeddings inside MLComputer isolate",
+        e,
+        s,
+      );
+      rethrow;
+    }
+  }
+
+  Future<void> cacheImageEmbeddings(List<EmbeddingVector> embeddings) async {
+    try {
+      await runInIsolate(
+        IsolateOperation.setIsolateCache,
+        {
+          'key': imageEmbeddingsKey,
+          'value': embeddings,
+        },
+      ) as bool;
+      _logger.info(
+        'Cached ${embeddings.length} image embeddings inside MLComputer',
+      );
+      return;
+    } catch (e, s) {
+      _logger.severe("Could not cache image embeddings in MLComputer", e, s);
+      rethrow;
+    }
+  }
+
+  Future<void> clearImageEmbeddingsCache() async {
+    try {
+      await runInIsolate(
+        IsolateOperation.clearIsolateCache,
+        {'key': imageEmbeddingsKey},
+      ) as bool;
+      return;
+    } catch (e, s) {
+      _logger.severe(
+        "Could not clear image embeddings cache in MLComputer",
+        e,
+        s,
+      );
+      rethrow;
+    }
+  }
 }
--- a/mobile/lib/services/machine_learning/ml_constants.dart
+++ b/mobile/lib/services/machine_learning/ml_constants.dart
@@ -0,0 +1 @@
+const imageEmbeddingsKey = "imageEmbeddings";
--- a/mobile/lib/services/machine_learning/semantic_search/query_result.dart
+++ b/mobile/lib/services/machine_learning/semantic_search/query_result.dart
@@ -0,0 +1,6 @@
+class QueryResult {
+  final int id;
+  final double score;
+
+  QueryResult(this.id, this.score);
+}
--- a/mobile/lib/services/machine_learning/semantic_search/semantic_search_service.dart
+++ b/mobile/lib/services/machine_learning/semantic_search/semantic_search_service.dart
@@ -1,12 +1,10 @@
-import "dart:async" show unawaited;
+import "dart:async" show Timer, unawaited;
 import "dart:developer" as dev show log;
 import "dart:math" show min;
 import "dart:ui" show Image;

-import "package:computer/computer.dart";
 import "package:flutter/foundation.dart";
 import "package:logging/logging.dart";
-import "package:ml_linalg/vector.dart";
 import "package:photos/core/cache/lru_map.dart";
 import "package:photos/core/event_bus.dart";
 import "package:photos/db/files_db.dart";
@@ -15,13 +13,14 @@ import 'package:photos/events/embedding_updated_event.dart';
 import "package:photos/models/file/file.dart";
 import "package:photos/models/ml/clip.dart";
 import "package:photos/models/ml/ml_versions.dart";
-import "package:photos/models/ml/vector.dart";
 import "package:photos/service_locator.dart";
 import "package:photos/services/collections_service.dart";
 import "package:photos/services/machine_learning/ml_computer.dart";
 import "package:photos/services/machine_learning/ml_result.dart";
 import "package:photos/services/machine_learning/semantic_search/clip/clip_image_encoder.dart";
+import "package:photos/services/machine_learning/semantic_search/query_result.dart";
 import "package:shared_preferences/shared_preferences.dart";
+import "package:synchronized/synchronized.dart";

 class SemanticSearchService {
  static final _logger = Logger("SemanticSearchService");
@@ -30,7 +29,6 @@ class SemanticSearchService {
  static final SemanticSearchService instance =
      SemanticSearchService._privateConstructor();

-  static final Computer _computer = Computer.shared();
  final LRUMap<String, List<double>> _queryEmbeddingCache = LRUMap(20);
  static const kMinimumSimilarityThreshold = 0.175;
  late final mlDataDB = MLDataDB.instance;
@@ -38,7 +36,11 @@ class SemanticSearchService {
  bool _hasInitialized = false;
  bool _textModelIsLoaded = false;

-  Future<List<EmbeddingVector>>? _cachedImageEmbeddingVectors;
+  final _cacheLock = Lock();
+  bool _imageEmbeddingsAreCached = false;
+  Timer? _embeddingsCacheTimer;
+  final Duration _embeddingsCacheDuration = const Duration(seconds: 60);
+
  Future<(String, List<EnteFile>)>? _searchScreenRequest;
  String? _latestPendingQuery;

@@ -53,12 +55,11 @@ class SemanticSearchService {
    _logger.info("init called");
    _hasInitialized = true;

-    // call getClipEmbeddings after 5 seconds
-    Future.delayed(const Duration(seconds: 5), () async {
-      await getClipVectors();
-    });
    Bus.instance.on<EmbeddingUpdatedEvent>().listen((event) {
-      _cachedImageEmbeddingVectors = null;
+      if (_imageEmbeddingsAreCached) {
+        MLComputer.instance.clearImageEmbeddingsCache();
+        _imageEmbeddingsAreCached = false;
+      }
    });

    unawaited(_loadTextModel(delay: true));
@@ -108,14 +109,21 @@ class SemanticSearchService {
    _logger.info("Indexes cleared");
  }

-  Future<List<EmbeddingVector>> getClipVectors() async {
-    if (_cachedImageEmbeddingVectors != null) {
-      return _cachedImageEmbeddingVectors!;
-    }
-    _cachedImageEmbeddingVectors ??= mlDataDB.getAllClipVectors();
-    _logger.info("read all embeddings from DB");
-
-    return _cachedImageEmbeddingVectors!;
+  Future<void> _cacheClipVectors() async {
+    return _cacheLock.synchronized(() async {
+      _resetInactivityTimer();
+      if (_imageEmbeddingsAreCached) {
+        return;
+      }
+      final now = DateTime.now();
+      final imageEmbeddings = await mlDataDB.getAllClipVectors();
+      _logger.info(
+        "read all ${imageEmbeddings.length} embeddings from DB in ${DateTime.now().difference(now).inMilliseconds} ms",
+      );
+      await MLComputer.instance.cacheImageEmbeddings(imageEmbeddings);
+      _imageEmbeddingsAreCached = true;
+      return;
+    });
  }

  Future<List<EnteFile>> getMatchingFiles(
@@ -257,17 +265,10 @@ class SemanticSearchService {
    required Map<String, double> minimumSimilarityMap,
  }) async {
    final startTime = DateTime.now();
-    final imageEmbeddings = await getClipVectors();
-    final Map<String, List<QueryResult>> queryResults = await _computer
-        .compute<Map<String, dynamic>, Map<String, List<QueryResult>>>(
-      computeBulkSimilarities,
-      param: {
-        "imageEmbeddings": imageEmbeddings,
-        "textQueryToEmbeddingMap": textQueryToEmbeddingMap,
-        "minimumSimilarityMap": minimumSimilarityMap,
-      },
-      taskName: "computeBulkSimilarities",
-    );
+    await _cacheClipVectors();
+    final Map<String, List<QueryResult>> queryResults = await MLComputer
+        .instance
+        .computeBulkSimilarities(textQueryToEmbeddingMap, minimumSimilarityMap);
    final endTime = DateTime.now();
    _logger.info(
      "computingSimilarities took for ${textQueryToEmbeddingMap.length} queries " +
@@ -278,6 +279,19 @@ class SemanticSearchService {
    return queryResults;
  }

+  void _resetInactivityTimer() {
+    _embeddingsCacheTimer?.cancel();
+    _embeddingsCacheTimer = Timer(_embeddingsCacheDuration, () {
+      _logger.info(
+        'Embeddings cache is unused for ${_embeddingsCacheDuration.inSeconds} seconds. Removing cache.',
+      );
+      if (_imageEmbeddingsAreCached) {
+        MLComputer.instance.clearImageEmbeddingsCache();
+        _imageEmbeddingsAreCached = false;
+      }
+    });
+  }
+
  static Future<ClipResult> runClipImage(
    int enteFileID,
    Image image,
@@ -296,50 +310,3 @@ class SemanticSearchService {
    return clipResult;
  }
 }
-
-Map<String, List<QueryResult>> computeBulkSimilarities(Map args) {
-  final imageEmbeddings = args["imageEmbeddings"] as List<EmbeddingVector>;
-  final textEmbedding =
-      args["textQueryToEmbeddingMap"] as Map<String, List<double>>;
-  final minimumSimilarityMap =
-      args["minimumSimilarityMap"] as Map<String, double>;
-  final result = <String, List<QueryResult>>{};
-  for (final MapEntry<String, List<double>> entry in textEmbedding.entries) {
-    final query = entry.key;
-    final textVector = Vector.fromList(entry.value);
-    final minimumSimilarity = minimumSimilarityMap[query]!;
-    final queryResults = <QueryResult>[];
-    if (!kDebugMode) {
-      for (final imageEmbedding in imageEmbeddings) {
-        final similarity = imageEmbedding.vector.dot(textVector);
-        if (similarity >= minimumSimilarity) {
-          queryResults.add(QueryResult(imageEmbedding.fileID, similarity));
-        }
-      }
-    } else {
-      double bestScore = 0.0;
-      for (final imageEmbedding in imageEmbeddings) {
-        final similarity = imageEmbedding.vector.dot(textVector);
-        if (similarity >= minimumSimilarity) {
-          queryResults.add(QueryResult(imageEmbedding.fileID, similarity));
-        }
-        if (similarity > bestScore) {
-          bestScore = similarity;
-        }
-      }
-      if (kDebugMode && queryResults.isEmpty) {
-        dev.log("No results found for query with best score: $bestScore");
-      }
-    }
-    queryResults.sort((first, second) => second.score.compareTo(first.score));
-    result[query] = queryResults;
-  }
-  return result;
-}
-
-class QueryResult {
-  final int id;
-  final double score;
-
-  QueryResult(this.id, this.score);
-}
--- a/mobile/lib/services/smart_memories_service.dart
+++ b/mobile/lib/services/smart_memories_service.dart
@@ -35,7 +35,6 @@ import "package:photos/services/location_service.dart";
 import "package:photos/services/machine_learning/face_ml/person/person_service.dart";
 import "package:photos/services/machine_learning/ml_computer.dart";
 import "package:photos/services/machine_learning/ml_result.dart";
-import "package:photos/services/machine_learning/semantic_search/semantic_search_service.dart";
 import "package:photos/services/search_service.dart";

 class MemoriesResult {
@@ -92,8 +91,7 @@ class SmartMemoriesService {
          await MLDataDB.instance.getFileIDsToFacesWithoutEmbedding();
      _logger.finest('fileIdToFaces has ${fileIdToFaces.length} entries $t');

-      final allImageEmbeddings =
-          await SemanticSearchService.instance.getClipVectors();
+      final allImageEmbeddings = await MLDataDB.instance.getAllClipVectors();
      _logger.finest(
        'allImageEmbeddings has ${allImageEmbeddings.length} entries $t',
      );
--- a/mobile/lib/utils/face/face_box_crop.dart
+++ b/mobile/lib/utils/face/face_box_crop.dart
@@ -10,7 +10,7 @@ import "package:photos/models/file/file.dart";
 import "package:photos/models/file/file_type.dart";
 import "package:photos/models/ml/face/box.dart";
 import "package:photos/models/ml/face/face.dart";
-import "package:photos/services/machine_learning/ml_computer.dart";
+import "package:photos/services/machine_learning/face_thumbnail_generator.dart";
 import "package:photos/utils/file_util.dart";
 import "package:photos/utils/thumbnail_util.dart";
 import "package:pool/pool.dart";
@@ -210,7 +210,7 @@ Future<Map<String, Uint8List>?> _getFaceCrops(
    faceBoxes.add(e.value);
  }
  final List<Uint8List> faceCrop =
-      await MLComputer.instance.generateFaceThumbnails(
+      await FaceThumbnailGenerator.instance.generateFaceThumbnails(
    // await generateJpgFaceThumbnails(
    imagePath,
    faceBoxes,
				`@@ -0,0 +1 @@`
				`const imageEmbeddingsKey = "imageEmbeddings";`