SwinStormer: High Resolution Image Deraining

We propose a Novel Deep Learning Architecture for the Image Deraining task, or in simple words, to remove rain from a single image.

Report Link: coming soon

Our architecture is inspired by the fusion of the following approaches:

• Restormer: Efficient Transformer for High-Resolution Image Restoration - https://arxiv.org/pdf/2111.09881.pdf
• Swin Transformer: Hierarchical Vision Transformer using Shifted Windows - https://arxiv.org/pdf/2103.14030

Requirements

pip install -r requirements.txt

Dataset

Rain100L and Rain100H are used, download these datasets and make ensure the directory structure looks like this:

|-- data     
    |-- rain100L
        |-- train
            |-- rain
                norain-1.png
                ...
            `-- norain
                norain-1.png
                ...
        `-- test                                                        
    |-- rain100H
        same as rain100L

Usage

You can easily train and test the model by running the command below. If you want to try other options, please refer to utils.py.

Train Model

The models were trained on one NVIDIA Tesla V100 GPU (8GB).

python main.py --data_name rain100L --seed 0

Test Model

Download the model from the link given at the bottom of the ReadMe or train the model before running this command:

python main.py --data_name rain100H --model_file result/rain100H.pth

Image Deraining Samples

The following images are taken from the model training phase. The improvement and deraining of the image sample can be observed.

Benchmarks

Model	Rain100L		Rain100H		Download
SwinStormer	PSNR	SSIM	PSNR	SSIM	Link
	38.68	0.981	29.13	0.868

License

This project is licensed under the terms of the GNU General Public License v3.0. You can find a copy of the license in the LICENSE file.