---
license: cc-by-nc-4.0
pipeline_tag: image-segmentation
library_name: pytorch
tags:
- medical-imaging
- foundation models
- fine-tuning
- retina
- fundus
- optic disc
- optic cup
- domain generalization
---

# FunduSegmenter

This is the official repository for **FunduSegmenter**, the first adaptation of the [RETFound](https://github.com/rmaphoh/RETFound) foundation model for joint optic disc (OD) and optic cup (OC) segmentation in retinal fundus images.

- **Paper:** [FunduSegmenter: Leveraging the RETFound Foundation Model for Joint Optic Disc and Optic Cup Segmentation in Retinal Fundus Images](https://huggingface.co/papers/2508.11354)
- **Code:** [GitHub - JusticeZzy/FunduSegmenter](https://github.com/JusticeZzy/FunduSegmenter)
- **License:** [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) (Non-commercial use only)

## Introduction
FunduSegmenter integrates a series of novel modules with RETFound, including a Pre-adapter, a Decoder, a Post-adapter, skip connections with Convolutional Block Attention Module (CBAM), and a Vision Transformer block adapter. The model demonstrates strong stability and generalization on both in-distribution and out-of-distribution data, providing stable OD and OC segmentation across multiple fundus camera datasets (IDRiD, Drishti-GS, RIM-ONE-r3, and REFUGE).

## Repository Contents
This repository hosts the pre-trained weights and datasets associated with the paper:

1. **Datasets.zip**: Processed datasets used in the study.
2. **DUNet_OD_CentreCrop_pretrained.pth**: Pre-trained DUNet used for OD center cropping pre-processing.
3. **FunduSegmenter_CroppedImage.pth**: Pre-trained FunduSegmenter trained on cropped ROI images.
4. **FunduSegmenter_OriginalImage.pth**: Pre-trained FunduSegmenter trained on original fundus images.
5. **Weights.zip**: All trained weights reported in the paper's experiments.
6. **mobilenet_v2-6a65762b.pth**: Required pre-trained weight for the DoFE baseline.

## License Note
While the code in the GitHub repository is under the MIT License, all the weights (including the original RETFound weights and the derived FunduSegmenter weights) are under **CC BY-NC 4.0** (Creative Commons Attribution-NonCommercial 4.0 International). This means the weights are for **non-commercial use only**.

## Citation
If you use our code or weights, please cite the following paper:

```bibtex
@article{zhao2025fundusegmenter,
  title={FunduSegmenter: Leveraging the RETFound Foundation Model for Joint Optic Disc and Optic Cup Segmentation in Retinal Fundus Images},
  author={Zhao, Zhenyi and Mookiah, M R K and Trucco, Emanuele},
  journal={arXiv preprint arXiv:2508.11354},
  year={2025}
}
```

## Contact
Zhenyi Zhao: zhenyi.zhao@hotmail.com or 2578745@dundee.ac.uk