pip install requirement.txt
- Download the original dataset from FF++.
- Extract frames from FF++ videos. The processed dataset can be downloaded from FF++ (code: 7s5s).
- Run the code in folder ./process to get the aligned images and masks.
Our model achieved the following performance on:
Training Data | Backbone | FF++ | Celeb-DF | DFDC_Pre | DFD |
---|---|---|---|---|---|
FF++ | Xception | 0.998 | 0.860 | 0.835 | 0.955 |
Note: the metric is frame-level AUC.
To train our model from scratch, please run :
python3 train.py --opt ./config/FF++.yml --gpu *
or
python -m torch.distributed.launch --nproc_per_node * --nnode * train.py --opt ./config/FF++.yml
@inproceedings{shuai2023locate,
title={Locate and Verify: A Two-Stream Network for Improved Deepfake Detection},
author={Shuai, Chao and Zhong, Jieming and Wu, Shuang and Lin, Feng and Wang, Zhibo and Ba, Zhongjie and Liu, Zhenguang and Cavallaro, Lorenzo and Ren, Kui},
booktitle={Proceedings of the 31st ACM International Conference on Multimedia},
pages={7131--7142},
year={2023}
}