An optic disk semantic segmentation method based on weakly supervised learning

Weakly supervised semantic segmentation has been widely used in the filed of computer vision. since it does not require expert annotations for training. Recently, some works use pseudo ground-truths which are generated by a classified network to train the model, however, this method is not suitable for medical image segmentation. To tackle this challenging problem, we use the GrabCut method to generate the coarse foreground segmentation map in this paper, and then we train the network based on a modified U-net model with the generated foreground map. Extensive experiments on the challenging RIM-ONE benchmarks strongly demonstrate the effectiveness of our algorithm. We obtain state-of-art results on RIM-ONE database.