[AZ-180] Enhance setup and improve inference logging

- Added a new Cython extension for the engine factory to the setup configuration.
- Updated the inference module to include additional logging for video batch processing and annotation callbacks.
- Refactored test cases to standardize the detection endpoint responses and include channel IDs in headers for better event handling.

src/engines/engine_factory.pyx

@@ -1,5 +1,6 @@
 import os
 import tempfile
+from loader_http_client cimport LoaderHttpClient, LoadResult
 
 
 class EngineFactory:
@@ -8,7 +9,9 @@ class EngineFactory:
     def create(self, model_bytes: bytes):
         raise NotImplementedError
 
-    def load_engine(self, loader_client, models_dir: str):
+    def load_engine(self, LoaderHttpClient loader_client, str models_dir):
+        cdef str filename
+        cdef LoadResult res
         filename = self._get_ai_engine_filename()
         if filename is None:
             return None
@@ -20,13 +23,13 @@ class EngineFactory:
             pass
         return None
 
-    def _get_ai_engine_filename(self) -> str | None:
+    def _get_ai_engine_filename(self):
         return None
 
-    def get_source_filename(self) -> str | None:
+    def get_source_filename(self):
         return None
 
-    def build_from_source(self, onnx_bytes: bytes, loader_client, models_dir: str):
+    def build_from_source(self, onnx_bytes, loader_client, models_dir):
         raise NotImplementedError(f"{type(self).__name__} does not support building from source")
 
 
@@ -35,7 +38,7 @@ class OnnxEngineFactory(EngineFactory):
         from engines.onnx_engine import OnnxEngine
         return OnnxEngine(model_bytes)
 
-    def get_source_filename(self) -> str:
+    def get_source_filename(self):
         import constants_inf
         return constants_inf.AI_ONNX_MODEL_FILE
 
@@ -45,7 +48,7 @@ class CoreMLEngineFactory(EngineFactory):
         from engines.coreml_engine import CoreMLEngine
         return CoreMLEngine(model_bytes)
 
-    def _get_ai_engine_filename(self) -> str:
+    def _get_ai_engine_filename(self):
         return "azaion_coreml.zip"
 
 
@@ -56,15 +59,15 @@ class TensorRTEngineFactory(EngineFactory):
         from engines.tensorrt_engine import TensorRTEngine
         return TensorRTEngine(model_bytes)
 
-    def _get_ai_engine_filename(self) -> str | None:
+    def _get_ai_engine_filename(self):
         from engines.tensorrt_engine import TensorRTEngine
         return TensorRTEngine.get_engine_filename()
 
-    def get_source_filename(self) -> str:
+    def get_source_filename(self):
         import constants_inf
         return constants_inf.AI_ONNX_MODEL_FILE
 
-    def build_from_source(self, onnx_bytes: bytes, loader_client, models_dir: str):
+    def build_from_source(self, onnx_bytes, loader_client, models_dir):
         from engines.tensorrt_engine import TensorRTEngine
         engine_bytes = TensorRTEngine.convert_from_source(onnx_bytes, None)
         return engine_bytes, TensorRTEngine.get_engine_filename()
@@ -75,11 +78,12 @@ class JetsonTensorRTEngineFactory(TensorRTEngineFactory):
         from engines.jetson_tensorrt_engine import JetsonTensorRTEngine
         return JetsonTensorRTEngine(model_bytes)
 
-    def _get_ai_engine_filename(self) -> str | None:
+    def _get_ai_engine_filename(self):
         from engines.tensorrt_engine import TensorRTEngine
         return TensorRTEngine.get_engine_filename("int8")
 
-    def build_from_source(self, onnx_bytes: bytes, loader_client, models_dir: str):
+    def build_from_source(self, onnx_bytes, LoaderHttpClient loader_client, str models_dir):
+        cdef str calib_cache_path
         from engines.tensorrt_engine import TensorRTEngine
         calib_cache_path = self._download_calib_cache(loader_client, models_dir)
         try:
@@ -92,10 +96,13 @@ class JetsonTensorRTEngineFactory(TensorRTEngineFactory):
                 except Exception:
                     pass
 
-    def _download_calib_cache(self, loader_client, models_dir: str) -> str | None:
+    def _download_calib_cache(self, LoaderHttpClient loader_client, str models_dir):
+        cdef LoadResult res
         import constants_inf
         try:
-            res = loader_client.load_big_small_resource(constants_inf.INT8_CALIB_CACHE_FILE, models_dir)
+            res = loader_client.load_big_small_resource(
+                constants_inf.INT8_CALIB_CACHE_FILE, models_dir
+            )
             if res.err is not None:
                 constants_inf.log(f"INT8 calibration cache not available: {res.err}")
                 return None

src/inference.pyx

@@ -268,14 +268,24 @@ cdef class Inference:
                 batch_count += 1
                 tf = total_frames if total_frames > 0 else max(frame_count, 1)
                 constants_inf.log(<str>f'Video batch {batch_count}: frame {frame_count}/{tf} ({frame_count*100//tf}%)')
+                last_ts = batch_timestamps[len(batch_timestamps) - 1] if batch_timestamps else 0
                 self._process_video_batch(ai_config, batch_frames, batch_timestamps, original_media_name, frame_count, tf, model_w)
+                if self._annotation_callback is not None:
+                    pann = Annotation(original_media_name, original_media_name, last_ts, [])
+                    cb = self._annotation_callback
+                    cb(pann, int(frame_count * 100 / tf))
                 batch_frames = []
                 batch_timestamps = []
         if batch_frames:
             batch_count += 1
             tf = total_frames if total_frames > 0 else max(frame_count, 1)
             constants_inf.log(<str>f'Video batch {batch_count} (flush): {len(batch_frames)} remaining frames')
+            last_ts = batch_timestamps[len(batch_timestamps) - 1] if batch_timestamps else 0
             self._process_video_batch(ai_config, batch_frames, batch_timestamps, original_media_name, frame_count, tf, model_w)
+            if self._annotation_callback is not None:
+                pann = Annotation(original_media_name, original_media_name, last_ts, [])
+                cb = self._annotation_callback
+                cb(pann, 100)
         constants_inf.log(<str>f'Video done: {frame_count} frames read, {batch_count} batches processed')
         self.send_detection_status()
 

src/main.py

@@ -645,6 +645,7 @@ async def detect_video_upload(
                     content_hash, _MEDIA_STATUS_AI_PROCESSED,
                     token_mgr.get_valid_token(),
                 )
+            await asyncio.sleep(0.01)
             _enqueue(channel_id, DetectionEvent(
                 annotations=[],
                 mediaId=content_hash,
@@ -681,8 +682,7 @@ async def detect_media(
     config: Annotated[Optional[AIConfigDto], Body()] = None,
     user_id: str = Depends(require_auth),
 ):
-    existing = _active_detections.get(media_id)
-    if existing is not None and not existing.done():
+    if media_id in _active_detections:
         raise HTTPException(status_code=409, detail="Detection already in progress for this media")
 
     channel_id = request.headers.get("x-channel-id", "")
@@ -779,7 +779,7 @@ async def detect_media(
             )
             _enqueue(channel_id, error_event)
         finally:
-            _active_detections.pop(media_id, None)
+            loop.call_later(5.0, lambda: _active_detections.pop(media_id, None))
             loop.call_later(10.0, _cleanup_channel, channel_id)
 
     _active_detections[media_id] = asyncio.create_task(run_detection())