CAIR-HKISI

CAIR-HKISI / SurgMotion

Public

Official Code for "SurgMotion: A Video-Native Foundation Model for Universal Understanding of Surgical Videos"

21
2
100% credibility
Found Mar 28, 2026 at 21 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

SurgMotion is an open-source framework for benchmarking AI foundation models on surgical video tasks such as phase recognition, action detection, and workflow analysis across multiple datasets.

How It Works

1
🔍 Discover SurgMotion

You stumble upon this helpful tool while looking for ways to make sense of surgery videos, promising smart insights into what doctors do during operations.

2
🛠️ Prepare your workspace

You quickly set up a simple space on your computer to handle video files and run tests, feeling ready to dive in.

3
📹 Gather surgery videos

You collect real surgery recordings from public sources and organize them into practice and test groups for analysis.

4
⬇️ Add smart understanding

You bring in ready-made brains trained on tons of surgery footage to recognize key moments and movements.

5
⚕️ Analyze surgery phases

With excitement, you watch the tool break down videos into steps like cutting, sewing, or checking, testing different surgery types.

6
📈 Check the insights

You review simple reports showing how accurately it spots actions and phases across various operations.

🏆 Unlock surgery smarts

You celebrate having top-notch understanding of surgical videos, ready to advance medical AI or research.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 21 to 21 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is SurgMotion?

SurgMotion is a Python framework for benchmarking video foundation models on surgical phase recognition, using 64-frame clips from nine datasets like Cholec80, AutoLaparo, and OphNet. It lets you download pre-trained SurgMotion models (300M or 1B params) from Hugging Face, prepare standardized data CSVs with prep scripts, and run linear probing via simple YAML configs and CLI commands like `python -m evals.main --fname config.yaml`. Developers get SOTA baselines comparing 15 models including DINOv3, EndoMamba, and EndoSSL in the official GitHub repository tied to a recent arXiv paper.

Why is it gaining traction?

It unifies evaluation across diverse surgical domains (laparoscopy, endoscopy, neurosurgery) with batch multi-GPU scripts and bootstrap sampling for robust metrics, saving weeks of setup. The flow-guided motion prediction handles surgery's textureless tissues better than pixel reconstruction, yielding top results on phase, action, and skill tasks. As the official code for SurgMotion-15M pretrained weights, it hooks researchers needing quick, reproducible surgical video benchmarks.

Who should use this?

ML engineers building surgical workflow predictors or robot assistants evaluating backbones like ViT or Mamba on real datasets. Researchers in medtech comparing endoscopy models for phase segmentation or skill scoring. Teams prototyping official codes for GI polyp detection or laparoscopic action recognition.

Verdict

Grab it for surgical video probing if you've got GPUs and datasets—setup is straightforward with conda/pip and clear README—but low 19 stars and 1.0% credibility score mean it's early; cross-check with the paper before production. Solid foundation for extension, not plug-and-play yet.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.