This is the code for paper: "Dual Defense: Adversarial, Traceable, and Invisible Robust Watermarking Against Face Swapping," in IEEE Transactions on Information Forensics and Security, vol. 19, pp. 4628-4641, 2024. https://ieeexplore.ieee.org/document/10486948
Our work proposes a novel active defense method Dual Defense based on robust adversarial watermark. Dual Defense embeds a robust adversarial watermark into the carrier facial image at one time, thereby destroying the deep face swapping model while tracking the copyright of the facial image.
You can download the faceswap model through the official link: https://github.com/Oldpan/Faceswap-Deepfake-Pytorch
The FaceNet model is used for performance evaluation and can be downloaded through the official link: https://github.com/davidsandberg/facenet
The core packages we use in the project and their version information are as follows:
kornia==0.6.12
numpy==1.22.3
pandas==1.3.5
torch==1.12.0
torchvision==0.13.0
python train.py
python test.py