This is an official PyTorch implementation of "Patch Diffusion: A General Module for Face Manipulation Detection" in AAAI2022.

conda create -n patch_diffusion python=3.6.10
conda install -y pytorch==1.4.0 torchvision==0.5.0 -c pytorch
pip install numpy==1.18.5

Dataset setup: Follow these instructions.

You can integrate pd module in your own network.

Pairwise Patch Loss (PPLoss) is to learn representative patch feature.

• CNN-generated images are surprisingly easy to spot...for now [Code]

• FaceForensics++: Learning to Detect Manipulated Facial Images [Code]

• DSP-FWA: Dual Spatial Pyramid for Exposing Face Warp Artifacts in DeepFake Videos [Code]

• kaggle-dfdc [Code]

If you use our code for your research, please cite the following paper:

@article{zhang2022pd,
  title={Patch Diffusion: A General Module for Face Manipulation Detection},
  author={Baogen Zhang, Sheng Li, Guorui Feng, Zhenxing Qian and Xinpeng Zhang},
  journal={AAAI},
  year={2022}
}