This is the installation, training and validating guide for RCDNet in the paper "A Model-driven Deep Neural Network for Single Image Rain Removal (CVPR2020)" Hong Wang, Qi Xie, Qian Zhao, and Deyu Meng [PDF] [Supplementary Material]
Run the script below to install (Conda):
conda create -n rcdnet python=3.6.7
conda activate rcdnet
conda install -y pytorch=0.4.1 cudatoolkit=9.0 torchvision -c pytorch
conda install h5py opencv
pip install scikit-image==0.17.2 pytorch-msssim==0.2.1 scipy==1.1.0
pip install ipython tdqm
To train and evaluate the models, please download training and testing datasets (Rain100H, Rain100L, Rain1400) from https://drive.google.com/file/d/1Q0hv7HQTT8iC5jHNb9lVrwI6yCNu1SI1/view?usp=sharing and place the unzipped folders into the 'for_syn/data' folder.
Configure the arguments in the train.sh and run the commands below:
$ cd ./RCDNet_code/for_syn/src/
$ bash train.sh
**Note that: For custom datasets, the image must have .png suffix
Since I don't want to change the orignal code so much, when you want to work on the Rain1400 dataset, you have to run convert_png.py and make_copies.py in the rain1400/train and rain1400/test directories.
Execute
$ cd ./RCDNet_code/for_syn/src/
$ bash test.sh
*Note that: --data_range 1-#training pairs/1-#testing pairs. The command (-- data_range 1-200/1-100) above is set based on Rain100L (200 training paris and 100 testing pairs).
The derained results are saved in the folder "./for_syn/experiment/...", where the image name "norain-*LR.png", "norain-*HR.png" , "norain-*SR.png" means rainy image, groundtruth, and restored background, respectively.