Update src/detection.py

src/detection.py

@@ -14,8 +14,12 @@ def _get_optimal_threads():
     physical_cores = psutil.cpu_count(logical=False)
     logical_cores = psutil.cpu_count(logical=True)
 
+    # Optimal intra-op threads = physical cores
+    # For high-performance scenarios, use physical cores
     intra_threads = physical_cores if physical_cores else 4
 
+    print(f"System info: {physical_cores} physical cores, {logical_cores} logical cores")
+    print(f"Using {intra_threads} intra-op threads for optimal performance")
 
     return intra_threads
 
@@ -45,7 +49,7 @@ class YOLOv11Detector:
         """Initialize YOLOv11 detector with maximum ONNX Runtime optimizations"""
         with open(config_path, 'r') as f:
             self.config = yaml.safe_load(f)
-        print(ort.get_device())
+
         model_path = self.config['model']['path']
 
         # Check which model file exists
@@ -79,25 +83,8 @@ class YOLOv11Detector:
     def _load_pytorch_model(self):
         """Load PyTorch model using Ultralytics"""
         from ultralytics import YOLO
-        import ultralytics.nn.modules as u_modules
-
-        # Patch Conv để tránh lỗi 'bn' lần đầu
-        if hasattr(u_modules, "Conv"):
-            old_init = u_modules.Conv.__init__
-
-            def new_init(self_, *args, **kwargs):
-                old_init(self_, *args, **kwargs)
-                if not hasattr(self_, "bn"):  # nếu thiếu bn thì gán None
-                    self_.bn = None
-
-            u_modules.Conv.__init__ = new_init
-
-        # Load YOLO model
         self.model = YOLO(self.model_path)
 
-        # Ngăn không cho gọi fuse() nữa
-        self.model.fuse = lambda *args, **kwargs: self.model
-
         # Set model to appropriate device
         if self.device == 'cuda:0' and torch.cuda.is_available():
             self.model.to('cuda')
@@ -186,6 +173,7 @@ class YOLOv11Detector:
         # For Intel CPUs: compact affinity for better cache usage
         os.environ['KMP_AFFINITY'] = 'granularity=fine,compact,1,0'
 
+        print(f"OpenMP configuration: threads={intra_threads}, policy=ACTIVE")
 
         # === CREATE OPTIMIZED SESSION ===
         self.session = ort.InferenceSession(
@@ -203,10 +191,15 @@ class YOLOv11Detector:
         try:
             # This might not always be available, but good to check
             model_meta = self.session.get_modelmeta()
+            print(f"Model metadata - Domain: {getattr(model_meta, 'domain', 'N/A')}")
         except:
             pass
 
         provider_used = self.session.get_providers()[0]
+        print(f"✅ ONNX Runtime v{ort.__version__} - Optimized session created")
+        print(f"📈 Provider: {provider_used}")
+        print(f"🧵 Threading: {intra_threads} intra-op threads, sequential execution")
+        print(f"🚀 Optimizations: Graph=ALL, Memory=Enabled, Spinning=Enabled, Dynamic=Enabled")
 
     def detect(self, image: np.ndarray) -> Dict:
         """
@@ -223,6 +216,7 @@ class YOLOv11Detector:
 
     def _detect_pytorch(self, image: np.ndarray) -> Dict:
         """Detection using PyTorch model"""
+        from ultralytics import YOLO
         results = self.model(
             image,
             conf=self.confidence,
@@ -316,7 +310,7 @@ class YOLOv11Detector:
             )
 
             if len(indices) > 0:
-                indices = np.array(indices).flatten()
+                indices = indices.flatten()
 
                 # Final results
                 final_boxes = [[int(x1[i]), int(y1[i]), int(x2[i]), int(y2[i])] for i in indices]