ZeroGPU compatible update

image_captioning.py

@@ -321,9 +321,9 @@ def analyze_video_activities(video_path):
         
         # Now process chunks
         chunk_generator = encode_video_in_chunks(video_path)
+        model, tokenizer, processor = load_model_and_tokenizer()
 
         for chunk_idx, video_frames, chunk_info in chunk_generator:
-            model, tokenizer, processor = load_model_and_tokenizer()
             prompt = "Analyze this construction site video chunk and describe the activities happening. Focus on construction activities, machinery usage, and worker actions. Include any construction equipment or machinery you can identify."
             response = process_video_chunk(video_frames, model, tokenizer, processor, prompt)
             print(f"Chunk {chunk_idx}: {response}")
@@ -364,11 +364,6 @@ def analyze_video_activities(video_path):
                 }
                 
                 all_activities.append(activity)
-            
-            # Cleanup
-            del model, tokenizer, processor
-            torch.cuda.empty_cache()
-            gc.collect()
         
         # Sort activities by timestamp
         all_activities.sort(key=lambda x: x['timestamp_seconds'])