Robust Real-World Image Super-Resolution against Adversarial Attacks-ACM MM 2021

Dr. Haofeng Li is recruiting students that want to pursue PhD/Master degrees in CUHK(SZ). Please check https://haofengli.net

This repository is an official PyTorch implementation of the paper "Robust Real-World Image Super-Resolution against Adversarial Attacks" from ACM MM 2021. If you have any problems about the code, please contact yuejt@mail2.sysu.edu.cn

We provide the training and testing codes, pre-trained models. You can train your model from scratch, or use the pre-trained model.

Code

Dependencies

Python 3.6
PyTorch >= 1.1.0
numpy
cv2
skimage
tqdm
torch-dct

Quick Start

Clone this github repo.

git clone https://github.com/yuejiutao/Robust-Real-World-Image-Super-Resolution-against-Adversarial-Attacks.git
cd Robust-Real-World-Image-Super-Resolution-against-Adversarial-Attacks

Folder

We recommend that you use the following directory structure：

yourfolder
└─Code
│   └─Robust-Real-World-Image-Super-Resolution-against-Adversarial-Attacks
│   └─Other project...
└─Data
│   └─RealSR
│       └─x4
│         └─test_LR
│         └─test_HR
│         └─train_LR
│         └─train_HR
│         └─adv
└─Other folder...

Training

Download the RealSR dataset(Version3) and unpack them like above. Then, change the dataroot and test_dataroot argument in ./options/realSR_HGSR_MSHR.py to the place where images are located.
Run the adversarial training with train.py using script file train.sh.

sh train.sh

You can change the exp_name in ./options/realSR_HGSR_MSHR.py and find the results in ./experiments/exp_name.

Testing

Download our pre-trained models to Robust-Real-World-Image-Super-Resolution-against-Adversarial-Attacks/pre/ folder or use your pre-trained models
Change the test_dataroot argument in test.sh to the place where images are located
Run test.sh.

sh test.sh

You can find the enlarged images under different adversarial intensities in /yourfolder/Data/RealSR/x4/adv/CDC_MC/ folder

Pretrained models

RobustRSR_X4

Citation

If you find our work useful in your research or publication, please cite:

@inproceedings{yue2021robust,
  title={Robust Real-World Image Super-Resolution against Adversarial Attacks},
  author={Yue, Jiutao and Li, Haofeng and Wei, Pengxu and Li, Guanbin and Lin, Liang},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
  pages={5148--5157},
  year={2021}
}

hubin111 / Robust-SR-against-Adversarial-Attacks