fix: limit each android background run to 20 mins

* fix: exif gps parsing of malformed data

* chore: e2e test

e2e/src/api/specs/asset.e2e-spec.ts

@@ -1140,6 +1140,16 @@ describe('/asset', () => {
           },
         },
       },
+      {
+        input: 'metadata/gps-position/empty_gps.jpg',
+        expected: {
+          type: AssetTypeEnum.Image,
+          exifInfo: {
+            latitude: null,
+            longitude: null,
+          },
+        },
+      },
     ];
 
     it.each(tests)(`should upload and generate a thumbnail for different file types`, async ({ input, expected }) => {

machine-learning/immich_ml/models/ocr/detection.py

@@ -1,8 +1,10 @@
 from typing import Any
 
+import cv2
 import numpy as np
+from numpy.typing import NDArray
 from PIL import Image
-from rapidocr.ch_ppocr_det import TextDetector as RapidTextDetector
+from rapidocr.ch_ppocr_det.utils import DBPostProcess
 from rapidocr.inference_engine.base import FileInfo, InferSession
 from rapidocr.utils import DownloadFile, DownloadFileInput
 from rapidocr.utils.typings import EngineType, LangDet, OCRVersion, TaskType
@@ -10,11 +12,10 @@ from rapidocr.utils.typings import ModelType as RapidModelType
 
 from immich_ml.config import log
 from immich_ml.models.base import InferenceModel
-from immich_ml.models.transforms import decode_cv2
 from immich_ml.schemas import ModelFormat, ModelSession, ModelTask, ModelType
 from immich_ml.sessions.ort import OrtSession
 
-from .schemas import OcrOptions, TextDetectionOutput
+from .schemas import TextDetectionOutput
 
 
 class TextDetector(InferenceModel):
@@ -24,13 +25,20 @@ class TextDetector(InferenceModel):
     def __init__(self, model_name: str, **model_kwargs: Any) -> None:
         super().__init__(model_name, **model_kwargs, model_format=ModelFormat.ONNX)
         self.max_resolution = 736
-        self.min_score = 0.5
-        self.score_mode = "fast"
+        self.mean = np.array([0.5, 0.5, 0.5], dtype=np.float32)
+        self.std_inv = np.float32(1.0) / (np.array([0.5, 0.5, 0.5], dtype=np.float32) * 255.0)
         self._empty: TextDetectionOutput = {
-            "image": np.empty(0, dtype=np.float32),
             "boxes": np.empty(0, dtype=np.float32),
             "scores": np.empty(0, dtype=np.float32),
         }
+        self.postprocess = DBPostProcess(
+            thresh=0.3,
+            box_thresh=model_kwargs.get("minScore", 0.5),
+            max_candidates=1000,
+            unclip_ratio=1.6,
+            use_dilation=True,
+            score_mode="fast",
+        )
 
     def _download(self) -> None:
         model_info = InferSession.get_model_url(
@@ -52,35 +60,65 @@ class TextDetector(InferenceModel):
 
     def _load(self) -> ModelSession:
         # TODO: support other runtime sessions
-        session = OrtSession(self.model_path)
-        self.model = RapidTextDetector(
-            OcrOptions(
-                session=session.session,
-                limit_side_len=self.max_resolution,
-                limit_type="min",
-                box_thresh=self.min_score,
-                score_mode=self.score_mode,
-            )
-        )
-        return session
+        return OrtSession(self.model_path)
 
-    def _predict(self, inputs: bytes | Image.Image) -> TextDetectionOutput:
-        results = self.model(decode_cv2(inputs))
-        if results.boxes is None or results.scores is None or results.img is None:
+    # partly adapted from RapidOCR
+    def _predict(self, inputs: Image.Image) -> TextDetectionOutput:
+        w, h = inputs.size
+        if w < 32 or h < 32:
+            return self._empty
+        out = self.session.run(None, {"x": self._transform(inputs)})[0]
+        boxes, scores = self.postprocess(out, (h, w))
+        if len(boxes) == 0:
             return self._empty
         return {
-            "image": results.img,
-            "boxes": np.array(results.boxes, dtype=np.float32),
-            "scores": np.array(results.scores, dtype=np.float32),
+            "boxes": self.sorted_boxes(boxes),
+            "scores": np.array(scores, dtype=np.float32),
         }
 
+    # adapted from RapidOCR
+    def _transform(self, img: Image.Image) -> NDArray[np.float32]:
+        if img.height < img.width:
+            ratio = float(self.max_resolution) / img.height
+        else:
+            ratio = float(self.max_resolution) / img.width
+
+        resize_h = int(img.height * ratio)
+        resize_w = int(img.width * ratio)
+
+        resize_h = int(round(resize_h / 32) * 32)
+        resize_w = int(round(resize_w / 32) * 32)
+        resized_img = img.resize((int(resize_w), int(resize_h)), resample=Image.Resampling.LANCZOS)
+
+        img_np: NDArray[np.float32] = cv2.cvtColor(np.array(resized_img, dtype=np.float32), cv2.COLOR_RGB2BGR)  # type: ignore
+        img_np -= self.mean
+        img_np *= self.std_inv
+        img_np = np.transpose(img_np, (2, 0, 1))
+        return np.expand_dims(img_np, axis=0)
+
+    def sorted_boxes(self, dt_boxes: NDArray[np.float32]) -> NDArray[np.float32]:
+        if len(dt_boxes) == 0:
+            return dt_boxes
+
+        # Sort by y, then identify lines, then sort by (line, x)
+        y_order = np.argsort(dt_boxes[:, 0, 1], kind="stable")
+        sorted_y = dt_boxes[y_order, 0, 1]
+
+        line_ids = np.empty(len(dt_boxes), dtype=np.int32)
+        line_ids[0] = 0
+        np.cumsum(np.abs(np.diff(sorted_y)) >= 10, out=line_ids[1:])
+
+        # Create composite sort key for final ordering
+        # Shift line_ids by large factor, add x for tie-breaking
+        sort_key = line_ids[y_order] * 1e6 + dt_boxes[y_order, 0, 0]
+        final_order = np.argsort(sort_key, kind="stable")
+        sorted_boxes: NDArray[np.float32] = dt_boxes[y_order[final_order]]
+        return sorted_boxes
+
     def configure(self, **kwargs: Any) -> None:
         if (max_resolution := kwargs.get("maxResolution")) is not None:
             self.max_resolution = max_resolution
-            self.model.limit_side_len = max_resolution
         if (min_score := kwargs.get("minScore")) is not None:
-            self.min_score = min_score
-            self.model.postprocess_op.box_thresh = min_score
+            self.postprocess.box_thresh = min_score
         if (score_mode := kwargs.get("scoreMode")) is not None:
-            self.score_mode = score_mode
-            self.model.postprocess_op.score_mode = score_mode
+            self.postprocess.score_mode = score_mode

machine-learning/immich_ml/models/ocr/recognition.py

@@ -1,9 +1,8 @@
 from typing import Any
 
-import cv2
 import numpy as np
 from numpy.typing import NDArray
-from PIL.Image import Image
+from PIL import Image
 from rapidocr.ch_ppocr_rec import TextRecInput
 from rapidocr.ch_ppocr_rec import TextRecognizer as RapidTextRecognizer
 from rapidocr.inference_engine.base import FileInfo, InferSession
@@ -14,6 +13,7 @@ from rapidocr.utils.vis_res import VisRes
 
 from immich_ml.config import log, settings
 from immich_ml.models.base import InferenceModel
+from immich_ml.models.transforms import pil_to_cv2
 from immich_ml.schemas import ModelFormat, ModelSession, ModelTask, ModelType
 from immich_ml.sessions.ort import OrtSession
 
@@ -65,17 +65,16 @@ class TextRecognizer(InferenceModel):
         )
         return session
 
-    def _predict(self, _: Image, texts: TextDetectionOutput) -> TextRecognitionOutput:
-        boxes, img, box_scores = texts["boxes"], texts["image"], texts["scores"]
+    def _predict(self, img: Image.Image, texts: TextDetectionOutput) -> TextRecognitionOutput:
+        boxes, box_scores = texts["boxes"], texts["scores"]
         if boxes.shape[0] == 0:
             return self._empty
         rec = self.model(TextRecInput(img=self.get_crop_img_list(img, boxes)))
         if rec.txts is None:
             return self._empty
 
-        height, width = img.shape[0:2]
-        boxes[:, :, 0] /= width
-        boxes[:, :, 1] /= height
+        boxes[:, :, 0] /= img.width
+        boxes[:, :, 1] /= img.height
 
         text_scores = np.array(rec.scores)
         valid_text_score_idx = text_scores > self.min_score
@@ -87,7 +86,7 @@ class TextRecognizer(InferenceModel):
             "textScore": text_scores[valid_text_score_idx],
         }
 
-    def get_crop_img_list(self, img: NDArray[np.float32], boxes: NDArray[np.float32]) -> list[NDArray[np.float32]]:
+    def get_crop_img_list(self, img: Image.Image, boxes: NDArray[np.float32]) -> list[NDArray[np.uint8]]:
         img_crop_width = np.maximum(
             np.linalg.norm(boxes[:, 1] - boxes[:, 0], axis=1), np.linalg.norm(boxes[:, 2] - boxes[:, 3], axis=1)
         ).astype(np.int32)
@@ -98,22 +97,55 @@ class TextRecognizer(InferenceModel):
         pts_std[:, 1:3, 0] = img_crop_width[:, None]
         pts_std[:, 2:4, 1] = img_crop_height[:, None]
 
-        img_crop_sizes = np.stack([img_crop_width, img_crop_height], axis=1).tolist()
-        imgs: list[NDArray[np.float32]] = []
-        for box, pts_std, dst_size in zip(list(boxes), list(pts_std), img_crop_sizes):
-            M = cv2.getPerspectiveTransform(box, pts_std)
-            dst_img: NDArray[np.float32] = cv2.warpPerspective(
-                img,
-                M,
-                dst_size,
-                borderMode=cv2.BORDER_REPLICATE,
-                flags=cv2.INTER_CUBIC,
-            )  # type: ignore
-            dst_height, dst_width = dst_img.shape[0:2]
+        img_crop_sizes = np.stack([img_crop_width, img_crop_height], axis=1)
+        all_coeffs = self._get_perspective_transform(pts_std, boxes)
+        imgs: list[NDArray[np.uint8]] = []
+        for coeffs, dst_size in zip(all_coeffs, img_crop_sizes):
+            dst_img = img.transform(
+                size=tuple(dst_size),
+                method=Image.Transform.PERSPECTIVE,
+                data=tuple(coeffs),
+                resample=Image.Resampling.BICUBIC,
+            )
+
+            dst_width, dst_height = dst_img.size
             if dst_height * 1.0 / dst_width >= 1.5:
-                dst_img = np.rot90(dst_img)
-            imgs.append(dst_img)
+                dst_img = dst_img.rotate(90, expand=True)
+            imgs.append(pil_to_cv2(dst_img))
+
         return imgs
 
+    def _get_perspective_transform(self, src: NDArray[np.float32], dst: NDArray[np.float32]) -> NDArray[np.float32]:
+        N = src.shape[0]
+        x, y = src[:, :, 0], src[:, :, 1]
+        u, v = dst[:, :, 0], dst[:, :, 1]
+        A = np.zeros((N, 8, 9), dtype=np.float32)
+
+        # Fill even rows (0, 2, 4, 6): [x, y, 1, 0, 0, 0, -u*x, -u*y, -u]
+        A[:, ::2, 0] = x
+        A[:, ::2, 1] = y
+        A[:, ::2, 2] = 1
+        A[:, ::2, 6] = -u * x
+        A[:, ::2, 7] = -u * y
+        A[:, ::2, 8] = -u
+
+        # Fill odd rows (1, 3, 5, 7): [0, 0, 0, x, y, 1, -v*x, -v*y, -v]
+        A[:, 1::2, 3] = x
+        A[:, 1::2, 4] = y
+        A[:, 1::2, 5] = 1
+        A[:, 1::2, 6] = -v * x
+        A[:, 1::2, 7] = -v * y
+        A[:, 1::2, 8] = -v
+
+        # Solve using SVD for all matrices at once
+        _, _, Vt = np.linalg.svd(A)
+        H = Vt[:, -1, :].reshape(N, 3, 3)
+        H = H / H[:, 2:3, 2:3]
+
+        # Extract the 8 coefficients for each transformation
+        return np.column_stack(
+            [H[:, 0, 0], H[:, 0, 1], H[:, 0, 2], H[:, 1, 0], H[:, 1, 1], H[:, 1, 2], H[:, 2, 0], H[:, 2, 1]]
+        )  # pyright: ignore[reportReturnType]
+
     def configure(self, **kwargs: Any) -> None:
         self.min_score = kwargs.get("minScore", self.min_score)

machine-learning/immich_ml/models/ocr/schemas.py

@@ -7,7 +7,6 @@ from typing_extensions import TypedDict
 
 
 class TextDetectionOutput(TypedDict):
-    image: npt.NDArray[np.float32]
     boxes: npt.NDArray[np.float32]
     scores: npt.NDArray[np.float32]
 

mobile/android/app/src/main/kotlin/app/alextran/immich/background/BackgroundEngineLock.kt

@@ -43,8 +43,8 @@ class BackgroundEngineLock(context: Context) : BackgroundWorkerLockApi, ImmichPl
 
   override fun onAttachedToEngine(binding: FlutterPlugin.FlutterPluginBinding) {
     super.onAttachedToEngine(binding)
-    checkAndEnforceBackgroundLock(binding.applicationContext)
     engineCount.incrementAndGet()
+    checkAndEnforceBackgroundLock(binding.applicationContext)
     Log.i(TAG, "Flutter engine attached. Attached Engines count: $engineCount")
   }
 

mobile/android/app/src/main/kotlin/app/alextran/immich/background/BackgroundWorker.g.kt

@@ -295,12 +295,12 @@ class BackgroundWorkerFlutterApi(private val binaryMessenger: BinaryMessenger, p
       } 
     }
   }
-  fun onAndroidUpload(callback: (Result<Unit>) -> Unit)
+  fun onAndroidUpload(maxMinutesArg: Long?, callback: (Result<Unit>) -> Unit)
 {
     val separatedMessageChannelSuffix = if (messageChannelSuffix.isNotEmpty()) ".$messageChannelSuffix" else ""
     val channelName = "dev.flutter.pigeon.immich_mobile.BackgroundWorkerFlutterApi.onAndroidUpload$separatedMessageChannelSuffix"
     val channel = BasicMessageChannel<Any?>(binaryMessenger, channelName, codec)
-    channel.send(null) {
+    channel.send(listOf(maxMinutesArg)) {
       if (it is List<*>) {
         if (it.size > 1) {
           callback(Result.failure(FlutterError(it[0] as String, it[1] as String, it[2] as String?)))

mobile/android/app/src/main/kotlin/app/alextran/immich/background/BackgroundWorker.kt

@@ -107,7 +107,7 @@ class BackgroundWorker(context: Context, params: WorkerParameters) :
    * This method acts as a bridge between the native Android background task system and Flutter.
    */
   override fun onInitialized() {
-    flutterApi?.onAndroidUpload { handleHostResult(it) }
+    flutterApi?.onAndroidUpload(maxMinutesArg = 20) { handleHostResult(it) }
   }
 
   // TODO: Move this to a separate NotificationManager class

mobile/android/app/src/main/kotlin/app/alextran/immich/background/BackgroundWorkerApiImpl.kt

@@ -5,8 +5,10 @@ import android.provider.MediaStore
 import android.util.Log
 import androidx.work.BackoffPolicy
 import androidx.work.Constraints
+import androidx.work.ExistingPeriodicWorkPolicy
 import androidx.work.ExistingWorkPolicy
-import androidx.work.OneTimeWorkRequest
+import androidx.work.OneTimeWorkRequestBuilder
+import androidx.work.PeriodicWorkRequestBuilder
 import androidx.work.WorkManager
 import io.flutter.embedding.engine.FlutterEngineCache
 import java.util.concurrent.TimeUnit
@@ -18,6 +20,7 @@ class BackgroundWorkerApiImpl(context: Context) : BackgroundWorkerFgHostApi {
 
   override fun enable() {
     enqueueMediaObserver(ctx)
+    enqueuePeriodicWorker(ctx)
   }
 
   override fun saveNotificationMessage(title: String, body: String) {
@@ -27,12 +30,14 @@ class BackgroundWorkerApiImpl(context: Context) : BackgroundWorkerFgHostApi {
   override fun configure(settings: BackgroundWorkerSettings) {
     BackgroundWorkerPreferences(ctx).updateSettings(settings)
     enqueueMediaObserver(ctx)
+    enqueuePeriodicWorker(ctx)
   }
 
   override fun disable() {
     WorkManager.getInstance(ctx).apply {
       cancelUniqueWork(OBSERVER_WORKER_NAME)
       cancelUniqueWork(BACKGROUND_WORKER_NAME)
+      cancelUniqueWork(PERIODIC_WORKER_NAME)
     }
     Log.i(TAG, "Cancelled background upload tasks")
   }
@@ -40,6 +45,7 @@ class BackgroundWorkerApiImpl(context: Context) : BackgroundWorkerFgHostApi {
   companion object {
     private const val BACKGROUND_WORKER_NAME = "immich/BackgroundWorkerV1"
     private const val OBSERVER_WORKER_NAME = "immich/MediaObserverV1"
+    private const val PERIODIC_WORKER_NAME = "immich/PeriodicBackgroundWorkerV1"
     const val ENGINE_CACHE_KEY = "immich::background_worker::engine"
 
 
@@ -55,7 +61,7 @@ class BackgroundWorkerApiImpl(context: Context) : BackgroundWorkerFgHostApi {
         setRequiresCharging(settings.requiresCharging)
       }.build()
 
-      val work = OneTimeWorkRequest.Builder(MediaObserver::class.java)
+      val work = OneTimeWorkRequestBuilder<MediaObserver>()
         .setConstraints(constraints)
         .build()
       WorkManager.getInstance(ctx)
@@ -67,10 +73,30 @@ class BackgroundWorkerApiImpl(context: Context) : BackgroundWorkerFgHostApi {
       )
     }
 
+    fun enqueuePeriodicWorker(ctx: Context) {
+      val settings = BackgroundWorkerPreferences(ctx).getSettings()
+      val constraints = Constraints.Builder().apply {
+        setRequiresCharging(settings.requiresCharging)
+      }.build()
+
+      val work =
+        PeriodicWorkRequestBuilder<PeriodicWorker>(
+          1,
+          TimeUnit.HOURS,
+          15,
+          TimeUnit.MINUTES
+        ).setConstraints(constraints)
+          .build()
+
+      WorkManager.getInstance(ctx)
+        .enqueueUniquePeriodicWork(PERIODIC_WORKER_NAME, ExistingPeriodicWorkPolicy.UPDATE, work)
+
+      Log.i(TAG, "Enqueued periodic background worker with name: $PERIODIC_WORKER_NAME")
+    }
+
     fun enqueueBackgroundWorker(ctx: Context) {
       val constraints = Constraints.Builder().setRequiresBatteryNotLow(true).build()
-
-      val work = OneTimeWorkRequest.Builder(BackgroundWorker::class.java)
+      val work = OneTimeWorkRequestBuilder<BackgroundWorker>()
         .setConstraints(constraints)
         .setBackoffCriteria(BackoffPolicy.EXPONENTIAL, 1, TimeUnit.MINUTES)
         .build()

mobile/android/app/src/main/kotlin/app/alextran/immich/background/PeriodicWorker.kt

@@ -0,0 +1,16 @@
+package app.alextran.immich.background
+
+import android.content.Context
+import android.util.Log
+import androidx.work.Worker
+import androidx.work.WorkerParameters
+
+class PeriodicWorker(context: Context, params: WorkerParameters) : Worker(context, params) {
+  private val ctx: Context = context.applicationContext
+
+  override fun doWork(): Result {
+    Log.i("PeriodicWorker", "Periodic worker triggered, starting background worker")
+    BackgroundWorkerApiImpl.enqueueBackgroundWorker(ctx)
+    return Result.success()
+  }
+}

mobile/ios/Runner/Background/BackgroundWorker.g.swift

@@ -295,7 +295,7 @@ class BackgroundWorkerBgHostApiSetup {
 /// Generated protocol from Pigeon that represents Flutter messages that can be called from Swift.
 protocol BackgroundWorkerFlutterApiProtocol {
   func onIosUpload(isRefresh isRefreshArg: Bool, maxSeconds maxSecondsArg: Int64?, completion: @escaping (Result<Void, PigeonError>) -> Void)
-  func onAndroidUpload(completion: @escaping (Result<Void, PigeonError>) -> Void)
+  func onAndroidUpload(maxMinutes maxMinutesArg: Int64?, completion: @escaping (Result<Void, PigeonError>) -> Void)
   func cancel(completion: @escaping (Result<Void, PigeonError>) -> Void)
 }
 class BackgroundWorkerFlutterApi: BackgroundWorkerFlutterApiProtocol {
@@ -326,10 +326,10 @@ class BackgroundWorkerFlutterApi: BackgroundWorkerFlutterApiProtocol {
       }
     }
   }
-  func onAndroidUpload(completion: @escaping (Result<Void, PigeonError>) -> Void) {
+  func onAndroidUpload(maxMinutes maxMinutesArg: Int64?, completion: @escaping (Result<Void, PigeonError>) -> Void) {
     let channelName: String = "dev.flutter.pigeon.immich_mobile.BackgroundWorkerFlutterApi.onAndroidUpload\(messageChannelSuffix)"
     let channel = FlutterBasicMessageChannel(name: channelName, binaryMessenger: binaryMessenger, codec: codec)
-    channel.sendMessage(nil) { response in
+    channel.sendMessage([maxMinutesArg] as [Any?]) { response in
       guard let listResponse = response as? [Any?] else {
         completion(.failure(createConnectionError(withChannelName: channelName)))
         return

mobile/lib/domain/services/background_worker.service.dart

@@ -122,46 +122,54 @@ class BackgroundWorkerBgService extends BackgroundWorkerFlutterApi {
   }
 
   @override
-  Future<void> onAndroidUpload() async {
-    _logger.info('Android background processing started');
-    final sw = Stopwatch()..start();
-    try {
-      if (!await _syncAssets(hashTimeout: Duration(minutes: _isBackupEnabled ? 3 : 6))) {
-        _logger.warning("Remote sync did not complete successfully, skipping backup");
-        return;
-      }
-      await _handleBackup();
-    } catch (error, stack) {
-      _logger.severe("Failed to complete Android background processing", error, stack);
-    } finally {
-      sw.stop();
-      _logger.info("Android background processing completed in ${sw.elapsed.inSeconds}s");
-      await _cleanup();
-    }
+  Future<void> onAndroidUpload(int? maxMinutes) async {
+    final hashTimeout = Duration(minutes: _isBackupEnabled ? 3 : 6);
+    final backupTimeout = maxMinutes != null ? Duration(minutes: maxMinutes - 1) : null;
+    return _backgroundLoop(
+      hashTimeout: hashTimeout,
+      backupTimeout: backupTimeout,
+      debugLabel: 'Android background upload',
+    );
   }
 
   @override
   Future<void> onIosUpload(bool isRefresh, int? maxSeconds) async {
-    _logger.info('iOS background upload started with maxSeconds: ${maxSeconds}s');
+    final hashTimeout = isRefresh ? const Duration(seconds: 5) : Duration(minutes: _isBackupEnabled ? 3 : 6);
+    final backupTimeout = maxSeconds != null ? Duration(seconds: maxSeconds - 1) : null;
+    return _backgroundLoop(hashTimeout: hashTimeout, backupTimeout: backupTimeout, debugLabel: 'iOS background upload');
+  }
+
+  Future<void> _backgroundLoop({
+    required Duration hashTimeout,
+    required Duration? backupTimeout,
+    required String debugLabel,
+  }) async {
+    _logger.info(
+      '$debugLabel started hashTimeout: ${hashTimeout.inSeconds}s, backupTimeout: ${backupTimeout?.inMinutes ?? '~'}m',
+    );
     final sw = Stopwatch()..start();
     try {
-      final timeout = isRefresh ? const Duration(seconds: 5) : Duration(minutes: _isBackupEnabled ? 3 : 6);
-      if (!await _syncAssets(hashTimeout: timeout)) {
+      if (!await _syncAssets(hashTimeout: hashTimeout)) {
         _logger.warning("Remote sync did not complete successfully, skipping backup");
         return;
       }
 
       final backupFuture = _handleBackup();
-      if (maxSeconds != null) {
-        await backupFuture.timeout(Duration(seconds: maxSeconds - 1), onTimeout: () {});
+      if (backupTimeout != null) {
+        await backupFuture.timeout(
+          backupTimeout,
+          onTimeout: () {
+            _cancellationToken.cancel();
+          },
+        );
       } else {
         await backupFuture;
       }
     } catch (error, stack) {
-      _logger.severe("Failed to complete iOS background upload", error, stack);
+      _logger.severe("Failed to complete $debugLabel", error, stack);
     } finally {
       sw.stop();
-      _logger.info("iOS background upload completed in ${sw.elapsed.inSeconds}s");
+      _logger.info("$debugLabel completed in ${sw.elapsed.inSeconds}s");
       await _cleanup();
     }
   }

mobile/lib/platform/background_worker_api.g.dart

@@ -273,7 +273,7 @@ abstract class BackgroundWorkerFlutterApi {
 
   Future<void> onIosUpload(bool isRefresh, int? maxSeconds);
 
-  Future<void> onAndroidUpload();
+  Future<void> onAndroidUpload(int? maxMinutes);
 
   Future<void> cancel();
 
@@ -327,8 +327,14 @@ abstract class BackgroundWorkerFlutterApi {
         pigeonVar_channel.setMessageHandler(null);
       } else {
         pigeonVar_channel.setMessageHandler((Object? message) async {
+          assert(
+            message != null,
+            'Argument for dev.flutter.pigeon.immich_mobile.BackgroundWorkerFlutterApi.onAndroidUpload was null.',
+          );
+          final List<Object?> args = (message as List<Object?>?)!;
+          final int? arg_maxMinutes = (args[0] as int?);
           try {
-            await api.onAndroidUpload();
+            await api.onAndroidUpload(arg_maxMinutes);
             return wrapResponse(empty: true);
           } on PlatformException catch (e) {
             return wrapResponse(error: e);

mobile/pigeon/background_worker_api.dart

@@ -47,7 +47,7 @@ abstract class BackgroundWorkerFlutterApi {
 
   // Android Only: Called when the Android background upload is triggered
   @async
-  void onAndroidUpload();
+  void onAndroidUpload(int? maxMinutes);
 
   @async
   void cancel();

server/src/services/metadata.service.ts

@@ -236,8 +236,8 @@ export class MetadataService extends BaseService {
       latitude: number | null = null,
       longitude: number | null = null;
     if (this.hasGeo(exifTags)) {
-      latitude = exifTags.GPSLatitude;
-      longitude = exifTags.GPSLongitude;
+      latitude = Number(exifTags.GPSLatitude);
+      longitude = Number(exifTags.GPSLongitude);
       if (reverseGeocoding.enabled) {
         geo = await this.mapRepository.reverseGeocode({ latitude, longitude });
       }
@@ -894,12 +894,10 @@ export class MetadataService extends BaseService {
     };
   }
 
-  private hasGeo(tags: ImmichTags): tags is ImmichTags & { GPSLatitude: number; GPSLongitude: number } {
-    return (
-      tags.GPSLatitude !== undefined &&
-      tags.GPSLongitude !== undefined &&
-      (tags.GPSLatitude !== 0 || tags.GPSLatitude !== 0)
-    );
+  private hasGeo(tags: ImmichTags) {
+    const lat = Number(tags.GPSLatitude);
+    const lng = Number(tags.GPSLongitude);
+    return !Number.isNaN(lat) && !Number.isNaN(lng) && (lat !== 0 || lng !== 0);
   }
 
   private getAutoStackId(tags: ImmichTags | null): string | null {

web/src/lib/components/shared-components/search-bar/search-bar.svelte

@@ -30,10 +30,10 @@
   let showSuggestions = $state(false);
   let isSearchSuggestions = $state(false);
   let selectedId: string | undefined = $state();
-  let isFocus = $state(false);
   let close: (() => Promise<void>) | undefined;
 
   const listboxId = generateId();
+  const searchTypeId = generateId();
 
   onDestroy(() => {
     searchStore.isSearchEnabled = false;
@@ -161,12 +161,10 @@
 
   const openDropdown = () => {
     showSuggestions = true;
-    isFocus = true;
   };
 
   const closeDropdown = () => {
     showSuggestions = false;
-    isFocus = false;
     searchHistoryBox?.clearSelection();
   };
 
@@ -251,6 +249,7 @@
         aria-activedescendant={selectedId ?? ''}
         aria-expanded={showSuggestions && isSearchSuggestions}
         aria-autocomplete="list"
+        aria-describedby={searchTypeId}
         use:shortcuts={[
           { shortcut: { key: 'Escape' }, onShortcut: onEscape },
           { shortcut: { ctrl: true, shift: true, key: 'k' }, onShortcut: onFilterClick },
@@ -287,12 +286,12 @@
       />
     </div>
 
-    {#if isFocus}
+    {#if searchStore.isSearchEnabled}
       <div
-        class="absolute inset-y-0 flex items-center"
+        id={searchTypeId}
+        class="absolute inset-y-0 flex items-center end-16"
         class:max-md:hidden={value}
-        class:end-16={isFocus}
-        class:end-28={isFocus && value.length > 0}
+        class:end-28={value.length > 0}
       >
         <p
           class="bg-immich-primary text-white dark:bg-immich-dark-primary/90 dark:text-black/75 rounded-full px-3 py-1 text-xs"