RETA

RETA benchmark contains color fundus images and pixel-level vessel annotations for retinal vessel segmentation and vascular tree analysis. Reserachers could develop and evaluate automated approaches using RETA. It aims to improve model generalization capability by learning on a dataset with less noisy labels.

Key Features:

(1) Images with presence of various imaging artefacts and retinal lesions.

(2) Annotations with less inter-annotator and intra-annotator variabilities.