we propose EfficientDerain for high-efficiency single-image deraining
- python 3.6
- pytorch 1.6.0
- opencv-python 4.4.0.44
- scikit-image 0.17.2
- torchvision 0.9.1
- pytorch-msssim 0.2.1
- Rain100L-old_version https://github.com/nnUyi/DerainZoo/blob/master/DerainDatasets.md
- Rain100H-old_version https://github.com/nnUyi/DerainZoo/blob/master/DerainDatasets.md
- Rain1400 https://xueyangfu.github.io/projects/cvpr2017.html
- SPA https://stevewongv.github.io/derain-project.html
Here is the urls of pretrained models (includes v3_rain100H, v3_rain1400, v3_SPA, v4_rain100H, v4_rain1400, v4_SPA) :
direct download: http://www.xujuefei.com/models_effderain.zip
google drive: https://drive.google.com/file/d/1OBAIG4su6vIPEimTX7PNuQTxZDjtCUD8/view?usp=sharing
baiduyun: https://pan.baidu.com/s/1kFWP-b3tD8Ms7VCBj9f1kw (pwd: vr3g)
- The code shown corresponds to version v3, for v4 change the value of argument "rainaug" in file "./train_*.sh" to the "true" (train_*.sh means it's the training script of dataset *)
- Unzip the "Streaks_Garg06.zip" in the "./rainmix"
- Change the value of argument "baseroot" in file "./train.sh" to the path of training data
- Edit the function "get_files" in file "./utils" according to the format of the training data
- Execute
sh train.sh
- The code shown corresponds to version v3
- Change the value of argument "load_name" in file "./test.sh" to the path of pretained model
- Change the value of argument "baseroot" in file "./test.sh" to the path of testing data
- Edit the function "get_files" in file "./utils" according to the format of the testing data
- Execute
sh test.sh
The specific results can be found in “./results/data/DERAIN.xlsx”
GT vs RCDNet |
GT vs EfDeRain |
Input vs GT |
GT vs RCDNet |
GT vs EfDeRain |
Input vs GT |
GT vs v1 |
GT vs v2 |
GT vs v3 |
GT vs v4 |
GT vs v1 |
GT vs v2 |
GT vs v3 |
GT vs v4 |
@inproceedings{guo2020efficientderain,
title={EfficientDeRain: Learning Pixel-wise Dilation Filtering for High-Efficiency Single-Image Deraining},
author={Qing Guo and Jingyang Sun and Felix Juefei-Xu and Lei Ma and Xiaofei Xie and Wei Feng and Yang Liu},
year={2021},
booktitle={AAAI}
}