feat: add test script for YOLO inference on surveillance video files

test_video.py

@@ -0,0 +1,54 @@
+import os
+from ultralytics import YOLO
+
+def main():
+    # 1. Paths
+    model_path = "person_detector_best.pt"
+    video_path = os.path.join(
+        "video_data", 
+        "S3 - 55 40 WJ -7", 
+        "download_ch18_20260406080008_20260406090008.mp4"
+    )
+    
+    # 2. Validations
+    if not os.path.exists(model_path):
+        print(f"[!] Error: Model not found at '{model_path}'")
+        return
+
+    if not os.path.exists(video_path):
+        print(f"[!] Error: Video not found at '{video_path}'")
+        return
+
+    # 3. Load Model
+    print(f"[*] Loading model: {model_path}")
+    # Using the Ultralytics YOLO class
+    model = YOLO(model_path)
+
+    # 4. Run Inference
+    print(f"[*] Running inference on video: {video_path}")
+    print("[*] Press 'q' in the video window to stop inference early.")
+    
+    # 'stream=True' is memory-efficient and processes the video frame by frame
+    # 'save=True' will save the annotated video to 'runs/detect/predict'
+    # 'show=True' will display the video as it processes
+    results = model.predict(
+        source=video_path,
+        conf=0.35,       # Confidence threshold (adjust if needed)
+        save=True,       # Save the output video
+        show=True,       # Show video in real-time
+        stream=True,      # Important for large 2GB video files
+        imgsz=1280,
+    )
+
+    # Since stream=True, we must iterate over the results generator to execute inference
+    try:
+        for _ in results:
+            pass
+    except KeyboardInterrupt:
+        print("\n[!] Inference interrupted by user.")
+        
+    print("\n[*] Inference finished.")
+    print("[*] Check the 'runs/detect/predict' directory (or similar) for your annotated video.")
+
+if __name__ == "__main__":
+    main()