Code for our ICASSP 2025 paper Convolutional Prompting for Broad-Domain Retinal Vessel Segmentation
Previous research on retinal vessel segmentation is targeted at a specific image domain, mostly color fundus photography (CFP). In this paper we make a brave attempt to attack a more challenging task of broad-domain retinal vessel segmentation (BD-RVS), which is to develop a unified model applicable to varied domains including CFP, SLO, UWF, OCTA and FFA. To that end, we propose Dual Convoltuional Prompting (DCP) that learns to extract domain-specific features by localized prompting along both position and channel dimensions. DCP is designed as a plug-in module that can effectively turn a R2AU-Net based vessel segmentation network to a unified model, yet without the need of modifying its network structure. For evaluation we build a broad-domain set using five public domain-specific datasets including ROSSA, FIVES, IOSTAR, PRIME-FP20 and VAMPIRE. In order to benchmark BD-RVS on the broad-domain dataset, we re-purpose a number of existing methods originally developed in other contexts, producing eight baseline methods in total. Extensive experiments show the the proposed method compares favorably against the baselines for BD-RVS.
| model | Resolution | Download | Mean AP | Mean AUC |
|---|---|---|---|---|
| UNet_DCP | 512*512 | HF | 0.6920 | 0.9736 |
| UNet_DCP | 1024*1024 | HF | 0.7862 | 0.9798 |
Install packages by pip install -r requirements.txt. This step is suggested to be done in your docker container or virtual environment or things like that.
Run python inference.py
If you find this our work useful, please consider citing:
@inproceedings{wei2025dcp,
title={Convolutional Prompting for Broad-Domain Retinal Vessel Segmentation},
author={Wei, Qijie and Yu, Weihong and Li, Xirong},
booktitle={ICASSP},
year={2025},
}
If you encounter any issue when running the code, please feel free to reach us either by creating a new issue in the GitHub or by emailing
- Qijie Wei (qijie.wei@ruc.edu.cn)