.

app.py

@@ -78,7 +78,8 @@ def process_video(
         add_watermark(temp_output, output_path)
         
         os.unlink(temp_output)
-        
+        progress(1.0, desc="Complete!")
+
         video_description = video_description[:500] + "..." if len(video_description) > 500 else video_description
         highlight_types = highlight_types[:500] + "..." if len(highlight_types) > 500 else highlight_types
         
@@ -125,7 +126,9 @@ def create_ui(examples_path: str):
                 label="Upload your video (max 20 minutes)",
                 interactive=True
             )
-            process_btn = gr.Button("Process Video", variant="primary")
+        
+        gr.Progress()
+        process_btn = gr.Button("Process Video", variant="primary")
 
         status = gr.Markdown(visible=True)
         
@@ -139,7 +142,7 @@ def create_ui(examples_path: str):
             output_video = gr.Video(label="Highlight Video", visible=False)
             download_btn = gr.Button("Download Highlights", visible=False)
 
-        def on_process(video):
+        def on_process(video, progress=gr.Progress()):
             if not video:
                 return {
                     status: "Please upload a video",
@@ -150,7 +153,7 @@ def create_ui(examples_path: str):
                 }
             
             status.value = "Processing video..."
-            output_path, desc, highlights, err = process_video(video)
+            output_path, desc, highlights, err = process_video(video, progress=progress)
             
             if err:
                 return {

video_highlight_detector.py

@@ -768,57 +768,3 @@ def load_model(
     
     return model, processor
 
-
-# def load_model(
-#     checkpoint_path: Optional[str] = None,
-#     base_model_id: str = "HuggingFaceTB/SmolVLM-2.2B-Instruct",
-#     device: str = "cuda"
-# ):
-#     """Load the model and processor."""
-#     # For demonstration, we set the target size
-#     video_target_size = 384
-
-#     processor = AutoProcessor.from_pretrained(base_model_id)
-#     # Configure the image processor
-#     processor.image_processor.size = {"longest_edge": video_target_size}
-#     processor.image_processor.do_resize = True
-#     processor.image_processor.do_image_splitting = False
-
-#     if checkpoint_path:
-#         model = SmolVLMForConditionalGeneration.from_pretrained(
-#             checkpoint_path,
-#             torch_dtype=torch.bfloat16,
-#             device_map=device
-#         )
-#     else:
-#         model =  SmolVLMForConditionalGeneration.from_pretrained(
-#             base_model_id,
-#             torch_dtype=torch.bfloat16,
-#             device_map=device
-#         )
-    
-#     return model, processor
-
-
-def main():
-    checkpoint_path = "/fsx/miquel/smolvlmvideo/checkpoints/final-visionUnfrozen-balanced/checkpoint-6550"
-    base_model_id = "HuggingFaceTB/SmolVLM-2.2B-Instruct"
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-
-    model, processor = load_model(checkpoint_path, base_model_id, device)
-    detector = BatchedVideoHighlightDetector(model, processor, device=device)
-
-    if len(sys.argv) < 3:
-        print("Usage: python video_highlight_detector.py <input_video> <output_video>")
-        sys.exit(1)
-
-    video_path = sys.argv[1]
-    output_path = sys.argv[2]
-
-    # Create highlight video
-    highlight_segments = detector.create_highlight_video(video_path, output_path)
-    print(f"Created highlight video with {len(highlight_segments)} segments")
-
-
-if __name__ == "__main__":
-    main()