Image Desnowing via Deep Invertible Separation

This is the official PyTorch implementation of the TCSVT 2023 paper.

Data

Put the training data you need under the directory 'data'.

Rename the directories that contains snowy images, clean images and snow masks to 'Snow', 'Gt' and 'Mask'.

Train

Set the training configs in train_config.py;

Set the root of the training data and cropping size in train.py by function Dataset();

If you want to continue training on a trained model, remember to reset the resume, resume_epoch and resume_optimizer in train_config.py.

Test

Set your traied model in test.py by function load_model();

Set the root of your testing data in test.py.

Pretrained model

We provide the pretrained model trained on CSD, and you could use it for testing. See the link below (pwd: invd):

Download Pretrained Model

Citation

@article{quan2023image,
  title={Image desnowing via deep invertible separation},
  author={Quan, Yuhui and Tan, Xiaoheng and Huang, Yan and Xu, Yong and Ji, Hui},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2023},
  publisher={IEEE}
}

Contacts

If you have questions, please contact with csxiaohengtan@foxmail.com

csxhtan / InvDSNet