Leminhbinh0209 / AAAI22-ADD

Official implementation of AAAI22 paper "ADD: Frequency Attention and Multi-View based Knowledge Distillation to Detect Low-Quality Compressed Deepfake Images"

ADD: Frequency Attention and Multi-View based Knowledge Distillation to Detect Low-Quality Compressed Deepfake Images

Thirty-Sixth AAAI Conference on Artificial Intelligence. 2022. Canada
[paper]

Overview of our framework

Installation

• Ubuntu 18.04.5 LTS
• CUDA 10.2
• Python 3.6.10
• python packages are detailed separately in requirements.txt.

Datasets

• NeuralTextures [Dataset] [Paper]
• DeepFakes [Dataset] [GitHub]
• Face2Face [Dataset] [Paper]
• FaceSwap [Dataset] [GitHub]
• FaceShifter [Dataset] [Paper]

Grad-CAM

Citation

Star (⭐) if you find it useful, and consider to cite our work

@inproceedings{le2022add,
  title={ADD: Frequency attention and multi-view based knowledge distillation to detect low-quality compressed deepfake images},
  author={Le, Binh M and Woo, Simon S},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={36},
  number={1},
  pages={122--130},
  year={2022}
}