SurgMotion Agent Prototype Assets

Frozen phase-prototype memories, offline feature bundles, and the SurgMotion-ONNX encoder for the SurgMotion doctor-demo agent.

Layout

manifest.json                          # onboard URLs + checksums (source of truth)
models/SurgMotion-ONNX/                # ViT-S/16 ONNX encoder (384-d patch tokens)
prototypes/<dataset>/surgmotion_vits/  # frozen phase prototype memory (npz + yaml)
features/<dataset>_surgmotion_vits_64f/ # optional offline 64-frame feature runs

Supported datasets in this bundle: Cholec80, PitVis.

Legacy paths (deprecated)

Older uploads used dinov3_vits in path names. Those objects may still exist for backward compatibility; new deployments should use surgmotion_vits paths listed in manifest.json.

Legacy path Canonical replacement
prototypes/<dataset>/dinov3_vits/ prototypes/<dataset>/surgmotion_vits/
features/<dataset>_dinov3_vits_64f/ features/<dataset>_surgmotion_vits_64f/

Upstream provenance

SurgMotion does not claim ownership of underlying ViT weights. The ONNX bundle is derived from publicly available exports; see models/SurgMotion-ONNX/README.md for upstream repository references and license notes.

Usage

Agent demos download assets via manifest.json (see demo/scripts/bootstrap_agent_demo.py or OpenCode onboard skill surgical-video-prototype-onboard).

export HF_TOKEN=hf_...
python demo/scripts/bootstrap_agent_demo.py

Maintainers refresh uploads with:

python demo/scripts/upload_agent_assets_to_hf.py --upload --update-skill
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support