Code for paper "Detecting Adversarial Faces Using Only Real Face Self-Perturbations".

Get Started

Datasets are LFW and CelebA-HQ.

Our codebase accesses the datasets from ./data/ and checkpoints from ./results/checkpoints/ by default.

├── ...
├── data
│   └── SP_new3
├── results
│   └──checkpoints
├── main.py
├── ...

Train Detector

sh SP_train.sh

Test

sh SP_test.sh

## Dependencies
python 3.8.8, PyTorch = 1.10.0, cudatoolkit = 11.7, torchvision, tqdm, scikit-learn, mmcv, numpy, opencv-python, dlib, Pillow

Datasets

LFW and Celeba-HQ datasets we used in this program are here. The generated adv-faces are also provided. You could generate adv-faces by torchattack. The attack code is in attack_utils.

Checkpoints

We provide some checkpoints for you to test. You can download them here. You can put them into the folder './results/checkpoints'.

To Test gradient-based adv-faces on LFW, run:

python main.py --config configs/datasets/SP_LFW.yml configs/pipelines/test/SP_test.yml --network.name X --network.checkpoint 'results/checkpoints/net-best_LFW.ckpt'

To Test gradient-based adv-faces on CelebA-HQ, run:

python main.py --config configs/datasets/SP_celebahq.yml configs/pipelines/test/SP_test.yml --network.name X --network.checkpoint 'results/checkpoints/net-best_celebahq.ckpt'

To Test GAN-based adv-faces on CelebA-HQ, run:

python main.py --config configs/datasets/GC_CA.yml configs/pipelines/test/SP_test.yml --network.name X_sep --network.checkpoint 'results/checkpoints/net-best_GC.ckpt'

Citation

If you find our repository useful for your research, please consider citing our paper:

@inproceedings{ijcai2023p165,
  title     = {Detecting Adversarial Faces Using Only Real Face Self-Perturbations},
  author    = {Wang, Qian and Xian, Yongqin and Ling, Hefei and Zhang, Jinyuan and Lin, Xiaorui and Li, Ping and Chen, Jiazhong and Yu, Ning},
  booktitle = {Proceedings of the Thirty-Second International Joint Conference on
               Artificial Intelligence, {IJCAI-23}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Edith Elkind},
  pages     = {1488--1496},
  year      = {2023},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2023/165},
  url       = {https://doi.org/10.24963/ijcai.2023/165},
}