This repository is the official PyTorch implementation of ARWGAN: attention-guided robust image watermarking model based on GAN.
The pre-trained model of ARWGAN is avaliable.
If you need to train ARWGAN from scratch, you should use commond line as following.
python mian.py new -n name -d data-dir -b batch-size -e epochs -n noise
Environmental requirements:
- Python == 3.7.4; Torch == 1.12.1 + cu102; Torchvision == 0.13.1; PIL == 7.2.0
Put the pre-trained model into pretrain floder, and you can test ARWGAN by command line as following.
python test.py -o ./pretrain/options-and-config.pickle -c ./pretrain/checkpoints/ARWGAN.pyt -s/mnt/chengxin/Datasets/DUTS/DUTS-TE/Std-Image-30/ -n 'Jpeg(10.0)'
@ARTICLE{10155247,
author={Huang, Jiangtao and Luo, Ting and Li, Li and Yang, Gaobo and Xu, Haiyong and Chang, Chin-Chen},
journal={IEEE Transactions on Instrumentation and Measurement},
title={ARWGAN: Attention-Guided Robust Image Watermarking Model Based on GAN},
year={2023},
volume={72},
number={},
pages={1-17},
doi={10.1109/TIM.2023.3285981}}The codes are designed based on HiDDeN.