Update app

app.py

@@ -1026,6 +1026,8 @@ def process_video_syncoffset(video_path, num_avg_frames, apply_preprocess):
 	except Exception as e:
 		return None, f"Error: {str(e)}"
 
+
+@spaces.GPU(duration=300)
 def process_video_activespeaker(video_path, global_speaker, num_avg_frames):
 	try:
 		# Extract the video filename

preprocess/inference_preprocess.py

@@ -33,6 +33,7 @@ parser.add_argument('--work_root', type=str, required=True, help='Path to save m
 parser.add_argument('--data_root', type=str, required=True, help='Directory containing ONLY full uncropped videos')
 opt = parser.parse_args()
 
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
 def bb_intersection_over_union(boxA, boxB):
 	xA = max(boxA[0], boxB[0])
@@ -184,7 +185,7 @@ def inference_video(opt, padding=0):
 			image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
 
 			# Perform person detection
-			results = yolo_model(image_np, verbose=False)
+			results = yolo_model(image_np.to(device), verbose=False)
 			detections = results[0].boxes
 
 			dets.append([])