Update chunk feature

app.py

@@ -126,7 +126,8 @@ def run_avse_inference(video_path, audio_path):
                 audio_chunk = np.concatenate([audio_chunk, pad])
 
             vid_idx = i // sr * 25  # convert audio index to video frame index
-            video_chunk = bg_frames[0, vid_idx:vid_idx+video_chunk_len, :, :]
+            video_chunk = bg_frames[vid_idx:vid_idx+video_chunk_len, :, :]
+            video_chunk = video_chunk[np.newaxis, ...]  # add batch dim
             if video_chunk.shape[0] < video_chunk_len:
                 pad = np.zeros((video_chunk_len - video_chunk.shape[0], *video_chunk.shape[1:]), dtype=video_chunk.dtype)
                 video_chunk = np.concatenate([video_chunk, pad], axis=0)