This is the implementation code for the paper titled: Restoring Snow-Degraded Single Images With Wavelet in Vision Transformer, IEEE Access 2023.

Images corrupted by snowy adverse weather can impose performance impediments to critical high-level vision-based applications. Restoring snow-degraded images is vital, but the task is ill-posed and very challenging due to the veiling effect, stochastic distribution, and multi-scale characteristics of snow in a scene. In this regard, many existing image denoising methods are often less successful with respect to snow removal, being that they mostly achieve success with one snow dataset and underperform in others, thus questioning their robustness in tackling real-world complex snowfall scenarios. In this paper, we propose the wavelet in transformer (WiT) network to address the image desnow inverse problem. Our model exploits the joint systemic capabilities of the vision transformer and the renowned discrete wavelet transform to achieve effective restoration of snow-degraded images. In our experiments, we evaluated the performance of our model on the popular SRRS, SNOW100K, and CSD datasets, respectively. The efficacy of our learning-based network is proven by our obtained numeric and qualitative result outcomes indicating significant performance gains compared to image desnow benchmark models and other state-of-the-art methods in the literature.

We recommend Python3 versions from 3.7 to 3.8, and CUDA versions from 10.2 to 11.5

• Download or Clone this repository and create environment

git clone https://github.com/obinnadyke/WiT
cd WiT

• Install the package dependencies pip3 install -r requirements.txt.

• PyTorch environment with GPU support for Windows and Linux can be installed with Conda conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch.

• Download the test-set images from the dataset links below and arrange in a data folder. Optionally, you can mogrify the image data formats. For example mogrify -format tif *.jpg && rm *.jpg to change .jpg images to .tif.

• Find the model_weights here and the dataset_filename.txt samples here.

• For inference, edit test.py accordingly and run python3 test.py -exp_name model_weight. For example python3 test.py -exp_name csd_weight for csd test dataset.

Create a data folder and arrange 2000 samples of the test input and groundtruth images as follows

    WiT
    |
    ├── data
    |   | 
    |   ├── test  # Test-set         
    |   |   ├── <dataset_name>          
    |   |   |   ├── input         # degraded images 
    |   |   |   └── gt            # clean images
    |   |   └── dataset_filename.txt

Use the link and download the respective snow-image datasets to evaluate the WiT network

• SNOW100K-L
• SRRS
• CSD

Qualitative PSNR and SSIM (numeric) results

Quantitative (visual) results evaluated on the CSD dataset

Bibtex:

@article{WiTNet2023,
  author={Obinna Agbodike and Jenhui Chen},
  journal={IEEE Access},
  title={Restoring Snow-Degraded Single Images With Wavelet in Vision Transformer},
  month=sep,
  year={2023},
  volume={11},
  pages={99470--99480}}

Useful blocks of code adapted in the WiT is credited to the contributions of ImageNetModel, TransWeather, and ViT-PyTorch