DIN

Introduction

This is the implementation of DIN base on Deep Interleaved Network for Image Super-Resolution With Asymmetric Co-Attention and Learning Deep Interleaved Networks with Asymmetric Co-Attention for Image Restoration. The Conference_version is for the former paper, which we take for example to edit the following instructions, and the Enhance_version is for the later. The architecture of our proposed DIN.

Environment

Python3
pytorch

Installations

skimage
imageio
tqdm
pandas
numpy
opencv-python
Matlab

Train

Download trainning set DIV2K and Flickr2K
Prepare training data. Run ./scripts/Prepare_TrainData_HR_LR.py or ./scripts/Prepare_TrainDate_HR_LR.m to generate HR/LR pairs with corresponding degradation models and scale factor. Modify ./scripts/flags.py to configure traindata_path and savedata_path.
Test data preparation is as same as train data preparation.
Configure ./options/train/train_DIN_x4.json for your training.
Run the command:

CUDA_VISIBLE_DEVICES=0 python train.py -opt options/train/train_DIN_x4.json

Test

Prepare testing data. Choose public standard benchmark datasets and run ./scripts/Prepare_TrainData_HR_LR.py or ./scripts/Prepare_TrainDate_HR_LR.m to generate HR/LR pairs with corresponding degradation models and scale factor. Modify ./scripts/flags.py to configure traindata_path and savedata_path.
Configure ./options/test/test_DIN_x4_BI.json for your testing.
Run the command and PSNR/SSIM values are printed and you can find the reconstructed images in ./result.

CUDA_VISIBLE_DEVICES=0 python test.py -opt options/test/test_DIN_x4_BI.json

Results

Here are the visual results.

Citation

If you find our work useful in your research or publications, please consider citing:

@inproceedings{dinsr,
author = {Li, Feng and Cong, Runmin and Bai, Huihui and He, Yifan},
title = {Deep Interleaved Network for Single Image Super-Resolution with Asymmetric Co-Attention},
booktitle = {International Joint Conference on Artificial Intelligence(ijcai)}
year = {2020},
month = {07},
pages = {537-543},
doi = {10.24963/ijcai.2020/75}
}

@article{dinir,
author = {Li, Feng and Cong, Runmin and Bai, Huihui and He, Yifan and Zhao, Yao and Zhu, Ce},
title = {Learning Deep Interleaved Networks with Asymmetric Co-Attention for Image Restoration}
booktitle = {arXiv:2010.15689}
year = {2020},
month = {10},
}

Acknowledgement

This code is built on SRFBN(Pytorch), we thank the authors for sharing their code.

lifengshiwo / DIN

DIN