by Xiaowei Hu, Yitong Jiang, Chi-Wing Fu, and Pheng-Ann Heng

This implementation is written by Xiaowei Hu at the Chinese University of Hong Kong.

Our USR dataset is available for download at Google Drive.

@inproceedings{hu2019mask,
     title={{Mask-ShadowGAN}: Learning to Remove Shadows from Unpaired Data},
     author={Hu, Xiaowei and Jiang, Yitong and Fu, Chi-Wing and Heng, Pheng-Ann},
     booktitle={ICCV},
     year={2019},
     note={to appear},
}

• Python 3.5
• PyTorch 1.0
• torchvision
• numpy

1. Select the training sets (USR, SRD, or ISTD ) and set the path of the dataset in train_Mask-ShadowGAN.py
2. Run train_Mask-ShadowGAN.py

1. Select the testing sets (USR, SRD, or ISTD ) and set the path of the dataset in test.py
2. Run test.py

Code is implemented based on a clean and readable Pytorch implementation of CycleGAN. We would like to thank Aitor Ruano and the authors of CycleGAN, Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros.