megvii-research / dcls-sr

Official PyTorch implementation of the paper "Deep Constrained Least Squares for Blind Image Super-Resolution", CVPR 2022.

[CVPR2022] Deep Constrained Least Squares for Blind Image Super-Resolution

Ziwei Luo¹, Haibin Huang², Lei Yu¹, Youwei Li¹, Haoqiang Fan¹, Shuaicheng Liu³

1. Megvii Research, 2. Kuaishou Technology

3. University of Electronic Science and Technology of China

This is the official implementation of 'Deep Constrained Least Squares for Blind Image Super-Resolution', CVPR 2022. [Paper]

Updates

[2022.04.22] 🎉🎉🎉 Our work BSRT won the 1st place in NTIRE 2022 BurstSR Challenge [Paper][Code].
[2022.03.09] We released the code and provided the pretrained model weights [here].
[2022.03.02] Our paper has been accepted by CVPR 2022.

Overview

Presentation Video:

[Youtube], [Bilibili]

Dependenices

• OS: Ubuntu 18.04
• nvidia :
  • cuda: 10.1
  • cudnn: 7.6.1
• python3
• pytorch >= 1.6
• Python packages: numpy opencv-python lmdb pyyaml

Dataset Preparation

We use DIV2K and Flickr2K as our training datasets (totally 3450 images).

To transform datasets to binary files for efficient IO, run:

python3 codes/scripts/create_lmdb.py

For evaluation of Isotropic Gaussian kernels (Gaussian8), we use five datasets, i.e., Set5, Set14, Urban100, BSD100 and Manga109.

To generate LRblur/LR/HR/Bicubic datasets paths, run:

python3 codes/scripts/generate_mod_blur_LR_bic.py

For evaluation of Anisotropic Gaussian kernels, we use DIV2KRK.

(You need to modify the file paths by yourself.)

Train

1. The core algorithm is in codes/config/DCLS.
2. Please modify codes/config/DCLS/options to set path, iterations, and other parameters...
3. To train the model(s) in the paper, run below commands.

For single GPU:

cd codes/config/DCLS
python3 train.py -opt=options/setting1/train_setting1_x4.yml

For distributed training

cd codes/config/DCLS
python3 -m torch.distributed.launch --nproc_per_node=4 --master_poer=4321 train.py -opt=options/setting1/train_setting1_x4.yml --launcher pytorch

Or choose training options use

cd codes/config/DCLS
sh demo.sh

Evaluation

To evalute our method, please modify the benchmark path and model path and run

cd codes/config/DCLS
python3 test.py -opt=options/setting1/test_setting1_x4.yml

Results

Comparison on Isotropic Gaussian kernels (Gaussian8)

Comparison on Anisotropic Gaussian kernels (DIV2KRK)

Citations

If our code helps your research or work, please consider citing our paper. The following is a BibTeX reference.

@inproceedings{luo2022deep,
  title={Deep constrained least squares for blind image super-resolution},
  author={Luo, Ziwei and Huang, Haibin and Yu, Lei and Li, Youwei and Fan, Haoqiang and Liu, Shuaicheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={17642--17652},
  year={2022}
}

Contact

email: [[email protected]]

Acknowledgement

This project is based on [DAN], [MMSR] and [BasicSR].