• Ubuntu 16.04, cuda 10.0
• Python 3.6.10, Pytorch 1.6.0
• More detail (See environment.yml)

1. Download the NTURain dataset from here or Baidu Cloud(Passwd:dtgv), and prepare the training data as follows:

   • Labled synthetic data:

         python makedata/preparedata_NTU.py  --ntu_path your_downloaded_synthetic_path --train_path your_saved_train_path 

   • Unlabled real data:

         python makedata/preparedata_NTU_semi.py  --ntu_path_semi your_downloaded_real_path --train_path your_saved_train_path

                Note that you should better put the synthetic and real training data sets into two different training folders.

1. Modify the configured file options_derain.json according to your own training and testing path.

2. Begin training:

       python main_NTURain.py
   

You need firstly download the testing dataset of NTURain and MSCSC into the folder testsets.

• NTURain synthetic data set:

      python test_NTURain_synthetic.py
  

  This manuscript will re-produce the paper results in Table 1.

• NTURain real data set:

      python test_NTURain_real.py
  

• MSCSC real data set:

      python test_MSCSC_real.py
  

@incollection{CVPR2021_2429,
title = {Semi-supervised video deraining with dynamical rain generator},
author = {Yue, Zongsheng and Xie, Jianwen and Zhao, Qian and Meng, Deyu},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year = {2021}
}