DF-RAP: A Robust Adversarial Perturbation for Defending against Deepfakes in Real-world Social Network Scenarios

Implementation of "DF-RAP: A Robust Adversarial Perturbation for Defending against Deepfakes in Real-world Social Network Scenarios".

Usage

You can follow demo.ipynb to implement robust adversarial attacks against Deepfakes.

Pretrained model

The pretrained model of ComGAN and PertG is available in ComGAN & PertG. Put them in DF-RAP/checkpoints/ .

The pretrained model of SimSwap and Arcface is available in SimSwap (old). Put them in DF-RAP/SimSwap/arcface_model/ and DF-RAP/SimSwap/checkpoints/ .

The pretrained model of StarGAN is available in StarGAN. Put it in DF-RAP/checkpoints/stargan_celeba_256/models/.

Dataset

We have made the OSN transmission image dataset mentioned in this work publicly available. You can get it here OSN-transmission mini CelebA.

Visualization

PGD-Based

Generator-Based

Citation

@article{qu2024df,
  title={DF-RAP: A Robust Adversarial Perturbation for Defending against Deepfakes in Real-world Social Network Scenarios},
  author={Qu, Zuomin and Xi, Zuping and Lu, Wei and Luo, Xiangyang and Wang, Qian and Li, Bin},
  journal={IEEE Transactions on Information Forensics and Security},
  year={2024},
  publisher={IEEE}
}

Acknowledges

Our work is based on:

[1] DiffJPEG

[2] StarGAN

[3] SimSwap

[4] Disrupting