SurgMotion-L

SurgMotion-L is the ViT-Large checkpoint of SurgMotion, a video-native foundation model for universal understanding of surgical videos.

This repository contains a single released checkpoint so that the model is structured as a standard single-model Hugging Face repository.

Files

• config.json: minimal repository config for Hub compatibility
• SurgMotion-vitl.safetensors: recommended weights in safetensors format
• SurgMotion-vitl.pt: original released PyTorch checkpoint retained for compatibility

Download

Download the repository config and safetensors checkpoint with huggingface_hub:

from huggingface_hub import hf_hub_download

config_path = hf_hub_download(
    repo_id="CAIR-HKISI/SurgMotion-vitl",
    filename="config.json",
)

weights_path = hf_hub_download(
    repo_id="CAIR-HKISI/SurgMotion-vitl",
    filename="SurgMotion-vitl.safetensors",
)

If your existing code expects the original PyTorch checkpoint format, keep using SurgMotion-vitl.pt.

Project Links

• Project page: https://surgmotion.cares-copilot.com/
• Paper: https://arxiv.org/abs/2602.05638
• Code: https://github.com/CAIR-HKISI/SurgMotion
• Full model collection: https://huggingface.co/CAIR-HKISI/SurgMotion