Datasets:

realslimman
/

REFUGE-MultiRater

language:
  - en
license: apache-2.0
size_categories:
  - 1K<n<10K
task_categories:
  - image-segmentation
  - image-classification
pretty_name: REFUGE
tags:
  - segmentation
  - fundus image
  - glaucoma
  - medical image

REFUGE

REFUGE Challenge provides a data set of 1200 fundus images with ground truth segmentations and clinical glaucoma labels, currently the largest existing one.

This dataset supplied multi-rater annotations of REFUGE Challenge Dataset. The challenge dataset releases majority vote (with some modifications) results of seven independent annotations. We release the scource seven annotations here.

Related Research

This dataset, REFUGE-MultiRater, is used as a challenging medical imaging benchmark in the paper Flow Stochastic Segmentation Networks. The paper introduces Flow Stochastic Segmentation Networks (Flow-SSN), a generative segmentation model family, achieving state-of-the-art results on this and other benchmarks.

The official code for this research can be found on GitHub.

Cite

@article{fang2022refuge2,
  title={REFUGE2 Challenge: Treasure for Multi-Domain Learning in Glaucoma Assessment},
  author={Fang, Huihui and Li, Fei and Wu, Junde and Fu, Huazhu and Sun, Xu and Cao, Xingxing and Son, Jaemin and Yu, Shuang and Zhang, Menglu and Yuan, Chenglang and Bian, Cheng and others},
  journal={arXiv preprint arXiv:2202.08994},
  year={2022},
  url={https://arxiv.org/abs/2202.08994}
}

@article{ribeiromedesign2025flow,
  title={Flow Stochastic Segmentation Networks},
  author={De Sousa Ribeiro, Fabio and Todd, Omar and Jones, Charles and Kori, Avinash and Mehta, Raghav and Glocker, Ben},
  journal={ICCV},
  year={2025},
  eprint={2507.18838},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}