DR-segmentation

Open-sourced code for the paper:

Improving Lesion Segmentation for Diabetic Retinopathy using Adversarial Learning, ICIAR, 2019

Please see our Presentation at Conference

Requirements:

pytorch 1.0, torchvision 0.2, numpy 1.16, scikit-learn 0.20, cv2 3.4, PIL 5.1 and ipdb 0.12.

Steps:

1. Download IDRiD dataset and put it under HEDNet_cGAN/data/.

2. To prepare for preprocessing, please run HEDNet_cGAN/blackmask.py to get the mask for each fundus image.

3. For training and evaluating UNet model, please go to UNet/. For training and evaluating HEDNet model, please go to HEDNet/. For training and evaluating HEDNet with conditional GAN model, please go to HEDNet_cGAN/.