Ml fixesv148 (#3028)

## Description

- Don't process ML on mobile for large files
- ML user developer options to clear local data

mobile/lib/db/ml/db.dart

@@ -668,6 +668,22 @@ class MLDataDB {
     return maps.first['count'] as int;
   }
 
+  Future<Set<int>> getErroredFileIDs() async {
+    final db = await instance.asyncDB;
+    final List<Map<String, dynamic>> maps = await db.getAll(
+      'SELECT DISTINCT $fileIDColumn FROM $facesTable WHERE $faceScore < 0',
+    );
+    return maps.map((e) => e[fileIDColumn] as int).toSet();
+  }
+
+  Future<void> deleteFaceIndexForFiles(List<int> fileIDs) async {
+    final db = await instance.asyncDB;
+    final String sql = '''
+      DELETE FROM $facesTable WHERE $fileIDColumn IN (${fileIDs.join(", ")})
+    ''';
+    await db.execute(sql);
+  }
+
   Future<int> getClusteredOrFacelessFileCount() async {
     final db = await instance.asyncDB;
     final List<Map<String, dynamic>> clustered = await db.getAll(

mobile/lib/models/ml/ml_versions.dart

@@ -7,3 +7,4 @@ const minimumClusterSize = 2;
 
 const embeddingFetchLimit = 200;
 final fileDownloadMlLimit = Platform.isIOS ? 5 : 10;
+const maxFileDownloadSize = 100000000;

mobile/lib/services/machine_learning/ml_indexing_isolate.dart

@@ -189,9 +189,8 @@ class MLIndexingIsolate {
   /// Analyzes the given image data by running the full pipeline for faces, using [_analyzeImageSync] in the isolate.
   Future<MLResult?> analyzeImage(
     FileMLInstruction instruction,
+    String filePath,
   ) async {
-    final String filePath = await getImagePathForML(instruction.file);
-
     final Stopwatch stopwatch = Stopwatch()..start();
     late MLResult result;
 

mobile/lib/services/machine_learning/ml_service.dart

@@ -391,8 +391,11 @@ class MLService {
     bool actuallyRanML = false;
 
     try {
+      final String filePath = await getImagePathForML(instruction.file);
+
       final MLResult? result = await MLIndexingIsolate.instance.analyzeImage(
         instruction,
+        filePath,
       );
       // Check if there's no result simply because MLController paused indexing
       if (result == null) {
@@ -467,25 +470,20 @@ class MLService {
       _logger.info("Results for file ${result.fileId} stored locally");
       return actuallyRanML;
     } catch (e, s) {
-      bool acceptedIssue = false;
       final String errorString = e.toString();
-      if (errorString.contains('ThumbnailRetrievalException')) {
-        _logger.severe(
-          'ThumbnailRetrievalException while processing image with ID ${instruction.file.uploadedFileID}, storing empty results so indexing does not get stuck',
-          e,
-          s,
-        );
-        acceptedIssue = true;
-      }
-      if (errorString.contains('InvalidImageFormatException')) {
-        _logger.severe(
-          '$errorString with ID ${instruction.file.uploadedFileID}, storing empty results so indexing does not get stuck',
-          e,
-          s,
-        );
-        acceptedIssue = true;
-      }
+      final String format = instruction.file.displayName.split('.').last;
+      final int? size = instruction.file.fileSize;
+      final fileType = instruction.file.fileType;
+      final bool acceptedIssue =
+          errorString.contains('ThumbnailRetrievalException') ||
+              errorString.contains('InvalidImageFormatException') ||
+              errorString.contains('FileSizeTooLargeForMobileIndexing');
       if (acceptedIssue) {
+        _logger.severe(
+          '$errorString with ID ${instruction.file.uploadedFileID} (format $format, type $fileType, size $size), storing empty results so indexing does not get stuck',
+          e,
+          s,
+        );
         await MLDataDB.instance.bulkInsertFaces(
           [Face.empty(instruction.file.uploadedFileID!, error: true)],
         );
@@ -495,7 +493,7 @@ class MLService {
         return true;
       }
       _logger.severe(
-        "Failed to analyze using FaceML for image with ID: ${instruction.file.uploadedFileID} and format ${instruction.file.displayName.split('.').last} (${instruction.file.fileType}). Not storing any results locally, which means it will be automatically retried later.",
+        "Failed to index file with ID: ${instruction.file.uploadedFileID} (format $format, type $fileType, size $size). Not storing any results locally, which means it will be automatically retried later.",
         e,
         s,
       );

mobile/lib/ui/settings/advanced_settings_screen.dart

@@ -13,7 +13,7 @@ import 'package:photos/ui/components/menu_item_widget/menu_item_widget.dart';
 import 'package:photos/ui/components/title_bar_title_widget.dart';
 import 'package:photos/ui/components/title_bar_widget.dart';
 import "package:photos/ui/components/toggle_switch_widget.dart";
-import "package:photos/ui/settings/machine_learning_settings_page.dart";
+import "package:photos/ui/settings/ml/machine_learning_settings_page.dart";
 import 'package:photos/ui/tools/debug/app_storage_viewer.dart';
 import 'package:photos/ui/viewer/gallery/photo_grid_size_picker_page.dart';
 import 'package:photos/utils/navigation_util.dart';

mobile/lib/ui/settings/ml/machine_learning_settings_page.dart

@@ -26,6 +26,7 @@ import "package:photos/ui/components/title_bar_title_widget.dart";
 import "package:photos/ui/components/title_bar_widget.dart";
 import "package:photos/ui/components/toggle_switch_widget.dart";
 import "package:photos/ui/settings/ml/enable_ml_consent.dart";
+import "package:photos/ui/settings/ml/ml_user_dev_screen.dart";
 import "package:photos/utils/ml_util.dart";
 import "package:photos/utils/network_util.dart";
 import "package:photos/utils/wakelock_util.dart";
@@ -42,6 +43,8 @@ class _MachineLearningSettingsPageState
     extends State<MachineLearningSettingsPage> {
   final EnteWakeLock _wakeLock = EnteWakeLock();
   Timer? _timer;
+  int _titleTapCount = 0;
+  Timer? _advancedOptionsTimer;
 
   @override
   void initState() {
@@ -55,6 +58,9 @@ class _MachineLearningSettingsPageState
         }
       });
     }
+    _advancedOptionsTimer = Timer.periodic(const Duration(seconds: 7), (timer) {
+      _titleTapCount = 0;
+    });
   }
 
   @override
@@ -63,6 +69,7 @@ class _MachineLearningSettingsPageState
     _wakeLock.disable();
     MachineLearningController.instance.forceOverrideML(turnOn: false);
     _timer?.cancel();
+    _advancedOptionsTimer?.cancel();
   }
 
   @override
@@ -73,8 +80,26 @@ class _MachineLearningSettingsPageState
         primary: false,
         slivers: <Widget>[
           TitleBarWidget(
-            flexibleSpaceTitle: TitleBarTitleWidget(
-              title: S.of(context).machineLearning,
+            flexibleSpaceTitle: GestureDetector(
+              child: TitleBarTitleWidget(
+                title: S.of(context).machineLearning,
+              ),
+              onTap: () {
+                setState(() {
+                  _titleTapCount++;
+                  if (_titleTapCount >= 7) {
+                    _titleTapCount = 0;
+                    // showShortToast(context, "Advanced options enabled");
+                    Navigator.of(context).push(
+                      MaterialPageRoute(
+                        builder: (BuildContext context) {
+                          return const MLUserDeveloperOptions();
+                        },
+                      ),
+                    ).ignore();
+                  }
+                });
+              },
             ),
             actionIcons: [
               IconButtonWidget(

mobile/lib/ui/settings/ml/ml_user_dev_screen.dart

@@ -0,0 +1,103 @@
+import "package:flutter/material.dart";
+import "package:photos/core/event_bus.dart";
+import "package:photos/db/ml/clip_db.dart";
+import "package:photos/db/ml/db.dart";
+import "package:photos/events/people_changed_event.dart";
+import "package:photos/services/machine_learning/semantic_search/semantic_search_service.dart";
+import "package:photos/theme/ente_theme.dart";
+import "package:photos/ui/components/buttons/button_widget.dart";
+import "package:photos/ui/components/models/button_type.dart";
+import "package:photos/ui/components/title_bar_title_widget.dart";
+import "package:photos/ui/components/title_bar_widget.dart";
+import "package:photos/utils/dialog_util.dart";
+import "package:photos/utils/toast_util.dart";
+
+class MLUserDeveloperOptions extends StatelessWidget {
+  const MLUserDeveloperOptions({super.key});
+
+  @override
+  Widget build(BuildContext context) {
+    return Scaffold(
+      body: CustomScrollView(
+        primary: false,
+        slivers: <Widget>[
+          const TitleBarWidget(
+            flexibleSpaceTitle: TitleBarTitleWidget(
+              title: "ML debug options",
+            ),
+          ),
+          SliverList(
+            delegate: SliverChildBuilderDelegate(
+              (delegateBuildContext, index) => Padding(
+                padding: const EdgeInsets.only(left: 16, right: 16),
+                child: Column(
+                  children: [
+                    Text(
+                      "Only use if you know what you're doing",
+                      textAlign: TextAlign.left,
+                      style: getEnteTextTheme(context).body.copyWith(
+                            color: getEnteColorScheme(context).textMuted,
+                          ),
+                    ),
+                    const SizedBox(height: 48),
+                    ButtonWidget(
+                      buttonType: ButtonType.neutral,
+                      labelText: "Purge empty indices",
+                      onTap: () async {
+                        await deleteEmptyIndices(context);
+                      },
+                    ),
+                    const SizedBox(height: 24),
+                    ButtonWidget(
+                      buttonType: ButtonType.neutral,
+                      labelText: "Reset all local ML",
+                      onTap: () async {
+                        await deleteAllLocalML(context);
+                      },
+                    ),
+                    const SafeArea(
+                      child: SizedBox(
+                        height: 12,
+                      ),
+                    ),
+                  ],
+                ),
+              ),
+              childCount: 1,
+            ),
+          ),
+        ],
+      ),
+    );
+  }
+
+  Future<void> deleteEmptyIndices(BuildContext context) async {
+    try {
+      final Set<int> emptyFileIDs = await MLDataDB.instance.getErroredFileIDs();
+      await MLDataDB.instance.deleteFaceIndexForFiles(emptyFileIDs.toList());
+      await MLDataDB.instance.deleteEmbeddings(emptyFileIDs.toList());
+      showShortToast(context, "Deleted ${emptyFileIDs.length} entries");
+    } catch (e) {
+      // ignore: unawaited_futures
+      showGenericErrorDialog(
+        context: context,
+        error: e,
+      );
+    }
+  }
+
+  Future<void> deleteAllLocalML(BuildContext context) async {
+    try {
+      await MLDataDB.instance.dropClustersAndPersonTable(faces: true);
+      await SemanticSearchService.instance.clearIndexes();
+      Bus.instance.fire(PeopleChangedEvent());
+      showShortToast(context, "All local ML cleared");
+    } catch (e) {
+      // ignore: unawaited_futures
+      showGenericErrorDialog(
+        context: context,
+        error: e,
+      );
+    }
+  }
+}

mobile/lib/ui/viewer/search_tab/people_section.dart

@@ -13,7 +13,7 @@ import "package:photos/models/search/search_constants.dart";
 import "package:photos/models/search/search_result.dart";
 import "package:photos/models/search/search_types.dart";
 import "package:photos/theme/ente_theme.dart";
-import "package:photos/ui/settings/machine_learning_settings_page.dart";
+import "package:photos/ui/settings/ml/machine_learning_settings_page.dart";
 import "package:photos/ui/viewer/file/no_thumbnail_widget.dart";
 import "package:photos/ui/viewer/file/thumbnail_widget.dart";
 import "package:photos/ui/viewer/people/add_person_action_sheet.dart";

mobile/lib/utils/ml_util.dart

@@ -153,7 +153,8 @@ Future<List<FileMLInstruction>> getFilesForMlIndexing() async {
 }
 
 Stream<List<FileMLInstruction>> fetchEmbeddingsAndInstructions(
-    int yieldSize,) async* {
+  int yieldSize,
+) async* {
   final List<FileMLInstruction> filesToIndex = await getFilesForMlIndexing();
   final List<List<FileMLInstruction>> chunks =
       filesToIndex.chunks(embeddingFetchLimit);
@@ -312,6 +313,12 @@ Future<String> getImagePathForML(EnteFile enteFile) async {
       throw ThumbnailRetrievalException(e.toString(), s);
     }
   } else {
+    // Don't process the file if it's too large (more than 100MB)
+    if (enteFile.fileSize != null && enteFile.fileSize! > maxFileDownloadSize) {
+      throw Exception(
+        "FileSizeTooLargeForMobileIndexing: size is ${enteFile.fileSize}",
+      );
+    }
     try {
       file = await getFile(enteFile, isOrigin: true);
     } catch (e, s) {