detect-fake-image

This is the source code of 'metric learning for anti-compression facial forgery detection'. We provide the codes and the trained models.

Data Preprocess

We experiment on FaceForensics++. Data partitioning can be found in jsons/. We use dlib to extract faces, and facial locations of frames are stored in jsons/. Use create_faces_from_dlib_locs.py to extract faces of every compression level and the corresponding mask area. The dataset folder should be

face
├── Deepfakes
│   ├── c23
│   │   ├── 000_003
│   │   │   ├── 105.png
│   │   │   ├── 106.png
│   │   │   ├── ...
│   │   │   ├── ...
│   │   ├── 001_870
│   │   ├── ...
│   │   ├── ...
│   ├── c40
│   └── masks
├── Face2Face
│   ├── c23
│   ├── c40
│   └── masks
├── FaceSwap
│   ├── c23
│   ├── c40
│   └── masks
├── NeuralTextures
│   ├── c23
│   ├── c40
│   └── masks
└── real
    ├── c23
    └── c40

Pretrained Models

Pretrained models using PyTorch are available at the link below. Baidu Drive: https://pan.baidu.com/s/1nfvPF-mSkD3FC8whpSpo8Q passcodes: 1234

Set up

To evlauate the performance of a pre-trained model, you should put the pretrained model released into ./checkpoints/ and and change DATASET in test.py and eval.py to select the test dataset. And then run

python test.py
python eval.py

Reference

Some of the codes are based on https://github.com/JStehouwer/FFD_CVPR2020 and https://github.com/ondyari/FaceForensics

Citation

@inproceedings{cao2021metric,
  title={Metric Learning for Anti-Compression Facial Forgery Detection},
  author={Cao, Shenhao and Zou, Qin and Mao, Xiuqing and Ye, Dengpan and Wang, Zhongyuan},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia (ACM MM 2021)},
  pages={1929--1937},
  year={2021}
}

If using the codes in your own work, you should also cite the following papers:

@inproceedings{cvpr2020-dang,
  title={On the Detection of Digital Face Manipulation},
  author={Hao Dang, Feng Liu, Joel Stehouwer, Xiaoming Liu, Anil Jain},
  booktitle={In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2020)},
  address={Seattle, WA},
  year={2020}
}

@inproceedings{roessler2019faceforensicspp,
	author = {Andreas R\"ossler and Davide Cozzolino and Luisa Verdoliva and Christian Riess and Justus Thies and Matthias Nie{\ss}ner},
	title = {Face{F}orensics++: Learning to Detect Manipulated Facial Images},
	booktitle= {International Conference on Computer Vision (ICCV)},
	year = {2019}
}

qinnzou / detect-fake-image