README.md

---
license: mit
tags:
  - medical-imaging
  - segmentation
  - retinal
  - pytorch
library_name: pytorch
inference: true
---

# RetSAM

<p align="left">
  <a href="https://arxiv.org/abs/2602.07012"><img src="https://img.shields.io/badge/arXiv-PDF-red" alt="arXiv"></a>
  <a href="https://wzhjerry.github.io/RetSAM/"><img src="https://img.shields.io/badge/Project-Page-blue" alt="Project Page"></a>
  <a href="https://huggingface.co/JerryWzh/RetSAM_public"><img src="https://img.shields.io/badge/Model-RetSAM-green" alt="Model"></a>
</p>


Overview
--------
Official repository for "A General Model for Retinal Segmentation and Quantification".

> **Note**: Publicly released models are trained on public datasets only and may differ from the paper-reported results.

Checkpoint
----------
- `--out_channels`: `"(2,3,2,4,6)"`.
- `--patch_size`: `4`.
- `--window_size`: `10`.
- `--depths`: `"(2, 2, 18, 2)"`.
- `--num_heads`: `"(4, 8, 16, 32)"`.
- `--feature_size`: `128`.
- `--size`: `640`.

Download from Hugging Face
--------------------------
Install `huggingface_hub`:
```bash
pip install -U huggingface_hub
```

Download the checkpoint with Python:
```python
from huggingface_hub import hf_hub_download

ckpt_path = hf_hub_download(
    repo_id="JerryWzh/RetSAM_public",
    filename="retsam_v1_for_public.ckpt",
)
print(ckpt_path)
```

Or download with CLI:
```bash
hf download JerryWzh/RetSAM_public retsam_v1_for_public.ckpt
```

Citation
--------
```bibtex
@article{wang2026general,
  title={A General Model for Retinal Segmentation and Quantification},
  author={Wang, Zhonghua and Ju, Lie and Li, Sijia and Feng, Wei and Zhou, Sijin and Hu, Ming and Xiong, Jianhao and Tang, Xiaoying and Peng, Yifan and Lin, Mingquan and others},
  journal={arXiv preprint arXiv:2602.07012},
  year={2026}
}
```

Contact
-------
For questions or access to full RetSAM models, please contact me at: zhonghua.wang@monash.edu