generate caption only once

app.py

@@ -89,16 +89,20 @@ def process_video(
     )
     result_file_name = generate_file_name(extension="mp4")
     result_file_path = os.path.join(RESULTS, result_file_name)
+
     TRACKER.reset()
+    caption = None
+
     with sv.VideoSink(result_file_path, video_info=video_info) as sink:
         for _ in tqdm(range(total // OUTPUT_LENGTH), desc="Processing video..."):
             frame = next(frame_generator)
-            caption = run_captioning(
-                model=MODEL,
-                processor=PROCESSOR,
-                image=frame,
-                device=DEVICE
-            )[CAPTIONING_TASK]
+            if caption is None:
+                caption = run_captioning(
+                    model=MODEL,
+                    processor=PROCESSOR,
+                    image=frame,
+                    device=DEVICE
+                )[CAPTIONING_TASK]
             detections = run_caption_to_phrase_grounding(
                 model=MODEL,
                 processor=PROCESSOR,