Mask-ShadowGAN: Learning to Remove Shadows from Unpaired Data

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.

---

USR Dataset

Our USR dataset is available for download at Google Drive.

Results

Please find the new results at https://github.com/xw-hu/Unveiling-Deep-Shadows.

Prerequisites

• Python 3.5
• PyTorch 1.0
• torchvision
• numpy

Train

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

Test

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

Bibtex

If you find our work, code, dataset, or results useful, please consider citing our paper as follows:

@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}
}

Acknowledgments

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.