Robust optic disc and cup segmentation with deep learning for glaucoma detection.

•Development and validation of the modified U-Net architecture with pre-trained ResNet model as encoder layers.•Realization of fully automated accurate segmentation of optic disc and cup for color fundus images, with segmentation performance comparable to that of experts for RIGA dataset.•Achievement of robust performance when applying the model trained on RIGA to DRISHTI-GS and RIM-ONE V3 database, indicating the robustness and generalization capability of the model.