Official PyTorch implementation of Diverse Generative Perturbations on Attention Space for Transferable Adversarial Attacks (ICIP 2022).

git clone https://github.com/wkim97/ADA.git
conda install --file requirements.txt

Download the training and evaluation datasets here and unzip the file under ADA/data.

The official evaluation dataset can also be downloaded from the NIPS 2017 adversarial attack competition.

You can download the pretrained weights here and unzip the file under ADA/weights.

python train.py --surrogate inception_v3 --target_layer Mixed_7c --save_dir ./weights --save_name default

python test.py --surrogate inception_v3 --target_layer Mixed_7c --load_dir ./weights --load_name default

Some parts of the code are borrowed from grad-cam-pytorch and from DSGAN.

If you find this code useful for your research, please consider citing our paper

@article{kim2022diverse,
  title={Diverse Generative Adversarial Perturbations on Attention Space for Transferable Adversarial Attacks},
  author={Kim, Woo Jae and Hong, Seunghoon and Yoon, Sung-Eui},
  journal={arXiv preprint arXiv:2208.05650},
  year={2022}
}