Domain-adaptive retinal vessel segmentation. A weakly supervised framework transferring vessel segmentation from DRIVE, CHASE_DB1 and FIVES to the unannotated RFMiD dataset, using confidence-masked pseudo-label supervision and perturbation consistency. Includes training, adaptation, evaluation and pseudo-label generation code.
deep-learning pytorch medical-imaging domain-adaptation retinal-vessel-segmentation fundus-images rfmid
-
Updated
Aug 5, 2026 - Jupyter Notebook