question about processing caption_sharegtpvideo_300k

Hi authors, thanks for your great work.
When using the codes for stage3 training, it seems the codes cannot process the catpion_sharegptvideo_300k dataset.

The "img" item in the following config (instruction_data.py) seems make no difference.

   caption_sharegptvideo_300k=[
        f"{my_path}/video/caption/sharegptvideo/train_300k.json", 
        "{my_path}/shareGPTVideo/train_video_and_instruction/train_300k/videos",
        "video",
        "img", # read from image
    ],

Thanks for your reply.

Start with the caption_sharegptvideo_300k entry in instruction_data.py and trace how stage3 training reads its JSON, video directory, modality, and "img" value. Compare that path with the reported dataset layout and determine why processing produces no effect; done means the dataset is handled as intended or the mismatch is documented with a reproducible result.