The code for "DATA AUGMENTATION IS MORE IMPORTANT THAN MODEL ARCHITECTURES FOR RETINAL VESSEL SEGMENTATION " We utilize the retina-unet "https://github.com/orobix/retina-unet" as the basic code, but we change a lot of code and add the rotation augmentation ,which achieves the good result AUC 0.9814.
To train the model,,
run bash train_1.sh