About use_multi_images_for_video

[yuezih]

I noticed that when setting use_multi_images_for_video to True, the input video frames seem to be repeated.

        if self.video_sampling_strategy.get('use_multi_images_for_video', False):
            new_frames = []
            for f in frames:
                new_frames.extend([f, f])
            frames = new_frames

As a result, setting n_frames to 16 leads to 32 images fed into the model.

Could you please tell me what the purpose of this operation is? Thanks a lot!