mismatch of sam2

[solitarytemple]

I was running the script visualize_patch_score_distribution.py and triggered a Hugging Face download for "--sam-model-id=facebook/sam2.1-hiera-small" when an error popped out: "You are using a model of type sam2_video to instantiate a model of type sam2". Then I went to the facebook repository of sam2.1-hiera-small and I found that the image and video tasks seem to share the same .pt weight file but the .json configuration differs which is video mode by default.

[ChengShiest]

Thanks for reporting this!

I checked facebook/sam2.1-hiera-small with transformers==4.56.2. This message is expected because the HF config.json has model_type: "sam2_video", while the script uses Sam2Model for image bbox-prompt segmentation.

I reproduced the warning, but the model still loads and bbox-prompt inference works. So this line is a warning, not necessarily the actual crash reason.

If the script exits, please share the full traceback after this warning..