---
title: SurgiTrack - Surgical Tool Tracking
emoji: 🔬
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
---

# SurgiTrack - Surgical Tool Tracking

Multi-class multi-tool tracking system for laparoscopic surgery videos.

## Overview

This demo implements the tracking pipeline from ["SurgiTrack: Fine-Grained Multi-Class Multi-Tool Tracking in Surgical Videos"](https://arxiv.org/abs/2312.07352), trained and evaluated on the CholecTrack20 dataset.

## Pipeline

1. **Detection**: YOLOv11x trained on 7 surgical tool classes
2. **Direction Estimation**: EfficientNet-B0 + Coordinate Attention predicts operator (MSLH, MSRH, ASRH)
3. **Tracking**: Operator-based slot assignment for graspers, fixed IDs for other tools

## Results

| Metric | Score |
|--------|-------|
| HOTA | 64.48% |
| AssA | 71.19% |
| DetA | 58.51% |

## Tool Classes

- Grasper (tracked by operator)
- Bipolar
- Hook
- Scissors
- Clipper
- Irrigator
- Specimen Bag

## Citation

```bibtex
@InProceedings{nwoye2023cholectrack20,
  author    = {Nwoye, Chinedu Innocent and Elgohary, Kareem and Srinivas, Anvita and Zaid, Fauzan and Lavanchy, Joël L. and Padoy, Nicolas},
  title     = {CholecTrack20: A Multi-Perspective Tracking Dataset for Surgical Tools},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2025},
  month     = {June}
}
```

## Author

[Djalil Khelladi](https://github.com/akhellad)