Suppress all debug logs in production mode

api_server.py

@@ -49,7 +49,7 @@ async def analyze_with_video(file: UploadFile = File(...), background_tasks: Bac
         
         # Tạo video có overlay
         output_video = os.path.join(output_dir, "analyzed_video.mp4")
-        reengineer_video(master_json, video_path, output_video)
+        reengineer_video(master_json, video_path, output_video, production=True)
         
         # Cleanup video gốc
         if os.path.exists(video_path): 

extract.py

@@ -25,7 +25,7 @@ def smooth_probs(probs, window_size=5):
     # Chia lại để tổng xác suất mỗi frame = 1
     return smoothed / smoothed.sum(axis=1, keepdims=True)
 
-def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_video=False, skip_phase_images=False, return_dict=False):
+def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_video=False, skip_phase_images=False, return_dict=False, production=False):
     if not os.path.exists(video_path):
         print(f"Lỗi: Không tìm thấy file {video_path}")
         return
@@ -55,11 +55,13 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
     save_dict = torch.load(model_path, map_location=device)
     model.load_state_dict(save_dict['model_state_dict'])
     model.to(device).eval()
-    print(f"Mô hình đã sẵn sàng trên {device} (Slow factor: {slow_factor})")
+    if not production:
+        print(f"Mô hình đã sẵn sàng trên {device} (Slow factor: {slow_factor})")
 
     transform = transforms.Compose([ToTensor(), Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
     
-    print(f"Đang đọc video {video_path}...")
+    if not production:
+        print(f"Đang đọc video {video_path}...")
     cap = cv2.VideoCapture(video_path)
     raw_frames = []
     while True:
@@ -74,7 +76,8 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
 
     # Frame Interpolation (Nội suy tuyến tính)
     if slow_factor < 1.0:
-        print(f"Đang nội suy frame (Slow-mo {slow_factor}x)...")
+        if not production:
+            print(f"Đang nội suy frame (Slow-mo {slow_factor}x)...")
         steps = int(1.0 / slow_factor)
         interpolated_full_res = []
         for i in range(len(raw_frames) - 1):
@@ -90,7 +93,8 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
         # Lưu file video slow-motion vật lý (optional)
         if not skip_slow_video:
             slow_video_path = os.path.join(output_dir, "slow_motion.mp4")
-            print(f"Đang ghi file slow motion: {slow_video_path}")
+            if not production:
+                print(f"Đang ghi file slow motion: {slow_video_path}")
             
             # Lấy lại FPS gốc (hoặc mặc định 30)
             cap = cv2.VideoCapture(video_path)
@@ -105,16 +109,18 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
             for frame in full_res_frames:
                 out.write(frame)
             out.release()
-            print(f"Đã tạo video slow motion: {slow_video_path}")
+            if not production:
+                print(f"Đã tạo video slow motion: {slow_video_path}")
         
     else:
         full_res_frames = raw_frames
 
     # Tiền xử lý cho AI
-    print("Đang tiền xử lý cho AI...")
+    if not production:
+        print("Đang tiền xử lý cho AI...")
     images = []
     input_size = 160
-    for img in tqdm(full_res_frames):
+    for img in tqdm(full_res_frames, disable=production):
         h, w = img.shape[:2]
         ratio = input_size / max(h, w)
         new_size = (int(w * ratio), int(h * ratio))
@@ -127,7 +133,8 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
     sample = transform({'images': np.asarray(images), 'labels': np.zeros(len(images))})
     img_tensor = sample['images'].unsqueeze(0).to(device)
     
-    print("Đang chạy AI Inference...")
+    if not production:
+        print("Đang chạy AI Inference...")
     with torch.no_grad():
         seq_length, batch, all_logits = 48, 0, []
         while batch * seq_length < img_tensor.shape[1]:
@@ -149,7 +156,8 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
         anchor_class = np.argmax(max_probs) # Class có độ tự tin cao nhất trong 8 sự kiện
         anchor_frame = np.argmax(probs_events[:, anchor_class])
         
-        print(f"Detected Anchor: {labels[anchor_class]} at frame {anchor_frame} (conf: {max_probs[anchor_class]:.2f})")
+        if not production:
+            print(f"Detected Anchor: {labels[anchor_class]} at frame {anchor_frame} (conf: {max_probs[anchor_class]:.2f})")
         
         events = np.zeros(8, dtype=int)
         events[anchor_class] = anchor_frame
@@ -184,7 +192,8 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
                 events[i] = total_frames - 1
             current_limit = events[i]
             
-        print(f"Detected Events (Frames): {events}")
+        if not production:
+            print(f"Detected Events (Frames): {events}")
 
         # Lưu thông tin frame index để visual_report sử dụng
         event_metadata = {}
@@ -205,13 +214,15 @@ def run_ai_extraction(video_path, slow_factor=1.0, output_dir=None, skip_slow_vi
         
         # Return dict nếu cần, hoặc ghi file (legacy)
         if return_dict:
-            print("Extraction complete (dict mode)")
+            if not production:
+                print("Extraction complete (dict mode)")
             return {"metadata": metadata}
         else:
             metadata_path = os.path.join(output_dir, "metadata.json")
             with open(metadata_path, 'w') as f:
                 json.dump(metadata, f)
-            print(f"Xong! Ảnh trích xuất lưu tại {phases_dir}")
+            if not production:
+                print(f"Xong! Ảnh trích xuất lưu tại {phases_dir}")
             return metadata
 
 if __name__ == "__main__":

main.py

@@ -40,7 +40,8 @@ def analyze_video_fast(video_path, production=True, output_file=None, output_bas
             output_dir=output_dir,
             skip_slow_video=True,  # Bỏ tạo slow-motion  
             skip_phase_images=False,  # CẦN ảnh phases để analyze.py phân tích
-            return_dict=True  # NEW: Return dict thay vì ghi file
+            return_dict=True,  # NEW: Return dict thay vì ghi file
+            production=production  # Tắt logs khi production=True
         )
         
         if not extraction_result:
@@ -90,7 +91,7 @@ def analyze_video_fast(video_path, production=True, output_file=None, output_bas
                     from reengineer import reengineer_video
                     output_video = os.path.join(os.path.dirname(output_file), 'analyzed_video.mp4')
                     print(f"\nĐang tạo video có overlay...")
-                    reengineer_video(output_file, video_path, output_video)
+                    reengineer_video(output_file, video_path, output_video, production=production)
                     print(f"Video đã lưu tại: {output_video}")
                 except Exception as e:
                     print(f"WARNING: Không thể tạo video overlay: {e}")

reengineer.py

@@ -12,8 +12,14 @@ if sys.stdout.encoding != 'utf-8':
     try:
         sys.stdout.reconfigure(encoding='utf-8')
     except AttributeError:
-        import io
-        sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8')
+        pass
+
+# Tắt logs từ TensorFlow/MediaPipe trong production
+import os as _os
+_os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'  # Tắt TF logs
+import logging
+logging.getLogger('tensorflow').setLevel(logging.ERROR)
+logging.getLogger('mediapipe').setLevel(logging.ERROR)
 
 POSE_CONNECTIONS = [
     (11, 12), (11, 13), (13, 15), (12, 14), (14, 16), # Vai - Khuỷu - Cổ tay
@@ -52,7 +58,7 @@ def draw_glass_panel(img, pt1, pt2, color=(0, 0, 0), alpha=0.5):
     cv2.rectangle(overlay, pt1, pt2, color, -1)
     return cv2.addWeighted(overlay, alpha, img, 1 - alpha, 0)
 
-def reengineer_video(json_path, video_path, output_path=None):
+def reengineer_video(json_path, video_path, output_path=None, production=False):
     """
     Áp dụng dữ liệu JSON lên video gốc (nguyên kích thước) với overlay cao cấp.
     """
@@ -98,7 +104,8 @@ def reengineer_video(json_path, video_path, output_path=None):
     else:
         temp_output_path = final_output_path
 
-    print(f"--- Đang tạo video Premium: {os.path.basename(video_path)} ({vw}x{vh}) ---")
+    if not production:
+        print(f"--- Đang tạo video Premium: {os.path.basename(video_path)} ({vw}x{vh}) ---")
 
     fourcc = cv2.VideoWriter_fourcc(*'mp4v')
     out = cv2.VideoWriter(temp_output_path, fourcc, fps_video, (vw, vh))
@@ -224,7 +231,8 @@ def reengineer_video(json_path, video_path, output_path=None):
             os.remove(video_path)
         os.rename(temp_output_path, final_output_path)
 
-    print(f"--- HOÀN TẤT! Video lưu tại: {final_output_path} ---")
+    if not production:
+        print(f"--- HOÀN TẤT! Video lưu tại: {final_output_path} ---")
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Golf Video Re-engineer Tool")