/UDL

PyTorch code for NeurIPS2021 paper "Uncertainty-Driven Loss for Single Image Super-Resolution"

Primary LanguagePython

Uncertainty-Driven Loss for Single Image Super-Resolution

[paper] [homepage]

This repository is Pytorch code for our proposed uncertainty-driven loss (UDL).

The code is built on RCAN and tested on Ubuntu 16.04 environment (Python 3.5/3.6/3.7, PyTorch 1.4.0) with 2080Ti/1080Ti GPUs.

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

@inproceedings{ning2021uncertainty,
  title={ Uncertainty-Driven Loss for Single Image Super-Resolution },
  author={ Ning Qian and Dong, WeiSheng and Li, Xin and Wu, Jinjian and Shi, Guangming },
  booktitle={Advances in Neural Information Processing Systems},
  year={2021}
}

Contents

  1. Requirements
  2. Test
  3. Acknowledgements

Requirements

  • Python 3
  • skimage
  • imageio
  • Pytorch (Pytorch version 1.0.1 is recommended)
  • tqdm
  • cv2 (pip install opencv-python)

Train

Quick start

   cd code
   sh train.sh

Test

Quick start

Test on standard SR benchmark

  1. If you have cloned this repository, the pre-trained models can be found in experiment fold and test dataset Set5 can be found in data fold.

  2. Then, run command:

    cd code
sh test.sh

  3. Finally, PSNR values are shown on your screen, you can find the reconstruction images in ../experiment/xx/results/

Acknowledgements

  • This code is built on RCAN (PyTorch). We thank the authors for sharing their codes.