Update app.py

app.py

@@ -64,7 +64,6 @@ def get_pipe_image_and_video_predictor():
     scheduler = UniPCMultistepScheduler.from_pretrained("./model/scheduler")
 
     pipe = Minimax_Remover_Pipeline(transformer=transformer, vae=vae, scheduler=scheduler)
-    pipe.to(device)
 
     sam2_checkpoint = "./SAM2-Video-Predictor/checkpoints/sam2_hiera_large.pt"
     config = "sam2_hiera_l.yaml"
@@ -244,7 +243,10 @@ def track_video(n_frames,video_state):
     video_state["origin_images"] = images
     images = np.array(images)
 
-    video_predictor_local=video_predictor.to("cuda")
+    sam2_checkpoint = "./SAM2-Video-Predictor/checkpoints/sam2_hiera_large.pt"
+    config = "sam2_hiera_l.yaml"
+    video_predictor_local = build_sam2_video_predictor(config, sam2_checkpoint, device="cuda")
+    
     inference_state = video_predictor_local.init_state(images=images/255, device="cuda")
     video_state["inference_state"] = inference_state
 
@@ -279,6 +281,7 @@ def track_video(n_frames,video_state):
         output_frames.append(painted)
     video_state["masks"] =mask_frames
     video_file = f"/tmp/{time.time()}-{random.random()}-tracked_output.mp4"
+    print("line 281 done")
     clip = ImageSequenceClip(output_frames, fps=15)
     clip.write_videofile(video_file, codec='libx264', audio=False, verbose=False, logger=None)
     return video_file,video_state