Awesome

DADA

Pytorch code of CVPR-2022 paper: Dual Adversarial Adaptation for Cross-Device Real-World Image Super-Resolution.

Requirements

• Pytorch == 1.5.1
• torchvision == 0.6.0
• opencv-python
• tensorboardX

Usage

Data Preparation

Download the DRealSR dataset and set the corresponding dataset path in constant.py.

Image files for different cameras are listed in ./image_list/train.csv and ./image_list/test.csv (for x4 scale factor).

Training

1. Train the up-sampling model:

   cd script
   sh train_up.sh
   

2. Train up-sampling model and down-sampling model collabotively:

   cd script
   sh train_circle.sh
   

3. Train DADA:

   cd script
   sh train_adapt.sh
   

Testing

cd script
sh test_adapt.sh

Pretrained models

Pretrianed models will be uploaded soon.

Citiation

@article{xu2022dual,
  title={Dual Adversarial Adaptation for Cross-Device Real-World Image Super-Resolution},
  author={Xu, Xiaoqian and Wei, Pengxu and Chen, Weikai and Mao, Mingzhi and Lin, Liang and Li, Guanbin},
  journal={arXiv preprint arXiv:2205.03524},
  year={2022}
}