This is the PyTorch implementation for our paper: Semi-MoreGAN: A Semi-supervised Image Mixture of Rain Removal Network (Pacific Graphics 2022) [(https://arxiv.org/abs/2204.13420)\].
- Python 3.6.0
- PyTorch (I use version 1.8.0. Suggest ≥ 1.2.0.)
- opencv
- numpy
- easydict
- skimage
Download the RainCityscapes training and validation images from Cityscapes website.
Organize the downloaded files as follows:
Semi-MoreGAN
├── datasets
│ ├── rain
│ ├── gt
│ ├── depth
│ ├── real
More details please see train_input.txt, train_gt.txt, train_depth.txt and train_real.txt.
Clone this repository:
git clone https://github.com/syy-whu/Semi-MoreGAN.git
run the supervised train code
python supervised_train.py
run the semi-supervised train code
python semi-supervised_train.py
Test the Semi-MoreGAN:
python3 infer.py
The PSNR and SSIM evaluation codes are from the skimage.
The code is based on DGNL-Net.
If you find this work useful in your research, please consider cite:
@inproceedings{shen2022semi,
title={Semi-MoreGAN: Semi-supervised Generative Adversarial Network for Mixture of Rain Removal},
author={Shen, Yiyang and Wang, Yongzhen and Wei, Mingqiang and Chen, Honghua and Xie, Haoran and Cheng, Gary and Wang, Fu Lee},
booktitle={Computer Graphics Forum},
volume={41},
number={7},
pages={443--454},
year={2022},
organization={Wiley Online Library}
}
@article{hu2021single,
title={Single-Image Real-Time Rain Removal Based on Depth-Guided Non-Local Features},
author={Hu, Xiaowei and Zhu, Lei and Wang, Tianyu and Fu, Chi-Wing and Heng, Pheng-Ann},
journal={IEEE Transactions on Image Processing},
volume={30},
pages={1759--1770},
year={2021},
publisher={IEEE}
}