DGNL-Net

Single-Image Real-Time Rain Removal Based on Depth-Guided Non-Local Features

by Xiaowei Hu, Lei Zhu, Tianyu Wang, Chi-Wing Fu, and Pheng-Ann Heng

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

Please find the code of the conference version at https://github.com/xw-hu/DAF-Net.

RainCityscapes Dataset

Our RainCityscapes dataset is available for download at the Cityscapes website.

Citations

@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}
}

@InProceedings{Hu_2019_CVPR,
     author = {Hu, Xiaowei and Fu, Chi-Wing and Zhu, Lei and Heng, Pheng-Ann},
     title = {Depth-Attentional Features for Single-Image Rain Removal},
     booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
     pages={8022--8031},
     year = {2019}
}

Prerequisites

Python 3.5
PyTorch 1.0

Installation

Clone this repository:

git clone https://github.com/xw-hu/DGNL-Net.git

Test

Please download our pretrained model at Google Drive.
Put the model 40000.pth in ./ckpt/DGNLNet/.
Put the model 60000.pth in ./ckpt/DGNLNet_fast/.

Test the DGNL-Net or DGNL-Net-fast:

python3 infer.py

python3 infer_fast.py

Train

Train the DGNL-Net model:
```
python3 train.py    
```
Train the DGNL-Net-fast model:
```
python3 train_fast.py
```

Evaluation

Please find the evaluation code at https://github.com/xw-hu/DAF-Net.
Enter the DAF-Net/examples/ and run evaluate_raincityscapes.m in Matlab.