iChallenge-GON is a large dataset containing 1200 annotated retinal fundus images.

The dataset was divided into three subsets at a ratio of 1:1:1, which were used for training, offline validation, and onsite testing. The proportion of glaucoma in each subset was the same.

The training set includes 400 color fundus images, which will provide the corresponding glaucoma status and unified manual pixel-wise annotation (aka ground truth). The test set consists of 800 color fundus images, of which 400 are off-site validation set images and 400 are on-site test set images.