remove resize frames process

video_cleaner.py

@@ -83,12 +83,9 @@ class VideoCleaner:
             prev_gray = gray
             
             # --- AI HUMAN DETECTION ---
-            # We only care about humans if the lights are on (as per your rules)
+            model = self._get_model()
-            # OR we can scan anyway if you want to find people in IR mode too.
+            # 直接传原图，省去 CPU resize，开启半精度和指定推理尺寸
-            # Following your rules: "If it's night and no lights, delete". 
+            results = model(frame, classes=[0], verbose=False, conf=0.7, half=True, imgsz=640)
-            # So we only need to detect humans to justify keeping the video.
-            small_frame = cv2.resize(frame, (640, 360))
-            results = self.model(small_frame, classes=[0], verbose=False, conf=0.7)
             debugmode=True
             if len(results[0].boxes) > 0:
                 if is_color:
@@ -156,28 +153,34 @@ class VideoCleaner:
         
         print(f"Scanning {input_path}...")
         
-        # 逐个目录递归处理，避免一次性加载所有文件到内存
+        all_videos = []
         for root, dirs, files in os.walk(input_path):
             if "processed" in dirs: 
                 dirs.remove("processed")
             
-            # 筛选当前目录下的视频文件
-            current_dir_videos = []
             for file in files:
                 if file.lower().endswith(self.supported_extensions):
-                    current_dir_videos.append(Path(root) / file)
+                    all_videos.append(Path(root) / file)
         
-            if not current_dir_videos:
+        if not all_videos:
-                continue
+            print("No videos found.")
+            return
 
-            # 对当前目录的文件按时间排序处理
+        # Sort by mtime
-            current_dir_videos.sort(key=lambda x: os.path.getmtime(x))
+        all_videos.sort(key=lambda x: os.path.getmtime(x))
         
-            print(f"Processing directory: {root} ({len(current_dir_videos)} files)")
+        print(f"Found {len(all_videos)} videos. Starting parallel processing with {self.workers} workers...")
-            for target_file in current_dir_videos:
+        
-                # 处理文件。如果 process_video_file 返回 False，
+        with ProcessPoolExecutor(max_workers=self.workers) as executor:
-                # 通常是因为文件太新（未到 30 天），则跳过当前文件
+            # We pass the method and its arguments. 
-                self.process_video_file(target_file, processed_dir, input_path)
+            # Note: self will be pickled. Since self.model is None, it should work.
+            futures = [executor.submit(self.process_video_file, v, processed_dir, input_path) for v in all_videos]
+            
+            for future in as_completed(futures):
+                try:
+                    future.result()
+                except Exception as e:
+                    print(f"Error in worker process: {e}")
 
 if __name__ == "__main__":
     import argparse
@@ -185,6 +188,7 @@ if __name__ == "__main__":
     parser.add_argument("dir", help="Target directory")
     parser.add_argument("--days", type=int, default=30)
     parser.add_argument("--model", type=str, default='yolo26n.pt', help="Path to YOLO model or model name")
+    parser.add_argument("--workers", type=int, default=4, help="Number of parallel workers")
     args = parser.parse_args()
-    cleaner = VideoCleaner(model_path=args.model, age_days=args.days)
+    cleaner = VideoCleaner(model_path=args.model, age_days=args.days, workers=args.workers)
     cleaner.scan_and_process(args.dir)