Semi-MoreGAN

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)\].

1. Requirements

Python 3.6.0
PyTorch (I use version 1.8.0. Suggest ≥ 1.2.0.)
opencv
numpy
easydict
skimage

2. Data preparation

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.

3. Main Training

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

4. Evaluation

Test the Semi-MoreGAN:

python3 infer.py

The PSNR and SSIM evaluation codes are from the skimage.

5. Acknowledgement

The code is based on DGNL-Net.

6. Citation

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}
}