/S2VD

Semi-supervised Video Deraining with Dynamical Rain Generator (CVPR, 2021, Pytorch)

Primary LanguagePythonMIT LicenseMIT

S2VD

Requirements and Dependencies

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

Training pipelines

  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
    

Testing pipelines

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
    

Citation

@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}
}