This repository is the official PyTorch implementation of HiFiMSFA: Robust and High-Fidelity Image Watermarking using Attention Augmented Deep Network. The paper proposes a deep image watermarking framework with multi-scale salient feature attention and adaptive squared Hinge loss.
If you need to train HiFiMSFA, you should use commond line as following.
python train.py
Requirements: torch == 1.10.1+cu111 torchvision == 0.11.2+cu111 kornia == 0.6.8
The pre-trained model of HiFiMSFA is avaliable at the pretrain floder. you can test it by command line as following.
python test.py
The models are free for non-commercial and scientific research purpose. Please mail us for further licensing terms.