# REFUGE2 Curated copy of the public dataset **REFUGE2**. Config `REFUGE2`. Other annotation files: config `REFUGE2-sidecar` (2 files). **Use and licence follow the original dataset.** Please cite the original publication and respect its terms. ## Original dataset | Field | Value | | :--- | :--- | | Country / region | China | | Acquisition centre | Zhongshan Ophthalmic Center, Sun Yat-sen University (glaucoma and myopia study cohort) | | Device | Zeiss Visucam 500 (2124×2056, training 400); Canon CR-2 (1634×1634, test 400); Kowa (1940×1940, validation 400; model not reported) | | Population | Chinese population; official public release of 1,200 images (of the 2,000 in the paper, the REFUGE validation set and the 400-image Topcon onsite set are not in the release) | | Original annotations | Optic disc / cup segmentation (1,200 images); glaucoma / non-glaucoma (training 40 / 360) and fovea coordinates (training 400; REFUGE annotations); validation and test have masks only | | Reference / link | Fang H, Li F, Wu J, et al. REFUGE2 Challenge: a treasure trove for multi-dimension analysis and evaluation in glaucoma screening. arXiv:2202.08994. 2022; REFUGE:Orlando JI, Fu H, Barbosa Breda J, et al. Med Image Anal. 2020;59:101570. doi:10.1016/j.media.2019.101570; https://refuge.grand-challenge.org | ## Contents - Images: 1,200 (train 395, validation 402, test 403); modalities: CFP 1,200. - Split rule: official (1,200). - Task labels: 2,400 rows; raster masks embedded: 1,200; other annotation files: 2. | Task | Label name | Rows | | :--- | :--- | ---: | | classification | glaucoma | 400 | | localization | Fovea_X | 400 | | localization | Fovea_Y | 400 | | segmentation | optic_disc_cup (0 bg, 1 disc rim, 2 cup) | 1,200 |