txmedai/vector-surgical-sam2-endovis18

Surgical SAM 2 — Vector package

Pinned Vector runtime package for the Surgical SAM 2 small checkpoint fine-tuned on EndoVis 2018 instrument masks. Its declared capability is prompted image segmentation: an RGB image and a bounding-box prompt produce an instrument mask and the model's predicted IoU. It is not an autonomous detector and not an RL policy.

Provenance:

• Upstream code and checkpoint: jinlab-imvr/Surgical-SAM-2 (Apache-2.0)
• Seldinger runtime fork: SeldingerMed/Surgical-SAM-2@fd424ac18126728d92125140f87374fdaf8ae78a
• Checkpoint SHA-256: c6202ee710badf747b4bf93fa3bfd8d82555e89d8ca3ff1e827493052d282955
• Training domain: EndoVis 2018
• Intended Vector evaluation domain: held-out EndoVis 2017 validation frames with box prompts derived from reference masks

The cross-dataset result is technical segmentation evidence only. It does not establish autonomous instrument detection, procedural safety, or clinical safety.