Uformer: A General U-Shaped Transformer for Image Restoration

Zhendong Wang, Xiaodong Cun, Jianmin Bao and Jianzhuang Liu

PWC PWC

Paper: https://arxiv.org/abs/2106.03106

Update:

  • 2021.08.19 Release a pre-trained model(Uformer32)! Add a script for FLOP/GMAC calculation.
  • 2021.07.29 Add a script for testing the pre-trained model on the arbitrary-resolution images.

In this paper, we present Uformer, an effective and efficient Transformer-based architecture, in which we build a hierarchical encoder-decoder network using the Transformer block for image restoration. Uformer has two core designs to make it suitable for this task. The first key element is a local-enhanced window Transformer block, where we use non-overlapping window-based self-attention to reduce the computational requirement and employ the depth-wise convolution in the feed-forward network to further improve its potential for capturing local context. The second key element is that we explore three skip-connection schemes to effectively deliver information from the encoder to the decoder. Powered by these two designs, Uformer enjoys a high capability for capturing useful dependencies for image restoration. Extensive experiments on several image restoration tasks demonstrate the superiority of Uformer, including image denoising, deraining, deblurring and demoireing. We expect that our work will encourage further research to explore Transformer-based architectures for low-level vision tasks.

Uformer

Details

Package dependencies

The project is built with PyTorch 1.7.1, Python3.7, CUDA10.1. For package dependencies, you can install them by:

pip3 install -r requirements.txt

Pretrained model

Data preparation

Denoising

For training data of SIDD, you can download the SIDD-Medium dataset from the official url. Then generate training patches for training by:

python3 generate_patches_SIDD.py --src_dir ../SIDD_Medium_Srgb/Data --tar_dir ../datasets/denoising/sidd/train

For evaluation, we use the same evaluation data as here, and put it into the dir ../datasets/denoising/sidd/val.

Training

Denoising

To train Uformer32(embed_dim=32) on SIDD, we use 2 V100 GPUs and run for 250 epochs:

python3 ./train.py --arch Uformer --batch_size 32 --gpu '0,1' \
    --train_ps 128 --train_dir ../datasets/denoising/sidd/train --env 32_0705_1 \
    --val_dir ../datasets/denoising/sidd/val --embed_dim 32 --warmup

More configuration can be founded in train.sh.

Evaluation

Denoising

To evaluate Uformer32 on SIDD, you can run:

python3 ./test.py --arch Uformer --batch_size 1 --gpu '0' \
    --input_dir ../datasets/denoising/sidd/val --result_dir YOUR_RESULT_DIR \
    --weights YOUR_PRETRAINED_MODEL_PATH --embed_dim 32 

Computational Cost

We provide a simple script to calculate the flops by ourselves, a simple script has been added in model.py. You can change the configuration and run it via:

python3 model.py

The manual calculation of GMacs in this repo differs slightly from the main paper, but they do not influence the conclusion. We will correct the paper later.

Citation

If you find this project useful in your research, please consider citing:

@article{wang2021uformer,
	title={Uformer: A General U-Shaped Transformer for Image Restoration},
	author={Wang, Zhendong and Cun, Xiaodong and Bao, Jianmin and Liu, Jianzhuang},
	journal={arXiv preprint 2106.03106},
	year={2021}
}

Acknowledgement

This code borrows heavily from MIRNet and SwinTransformer.

Contact

Please contact us if there is any question or suggestion(Zhendong Wang , Xiaodong Cun [email protected]).

GitHub

https://github.com/ZhendongWang6/Uformer