opencompass/MMBench-Video · more code examples

Could you provide more code examples on how to use this benchmark, instead of just a single line like:
python run.py --model GPT4o --data MMBench-Video --nframe 8 --verbose?

VLMEvalKit is too large and complex to easily follow. For example, how do I specify the path to a local dataset? I’ve searched for a long time and still couldn’t find an answer. It seems to point to this file:
https://github.com/open-compass/VLMEvalKit/blob/main/vlmeval/dataset/video_dataset_config.py,
but there are no clear instructions on how to modify it.

So, if I’ve downloaded those .pkl files, how am I supposed to use them? It’s really confusing.