Offical implemention of Robust Superpixel-Guided Attentional Adversarial Attack (CVPR2020)


  • Install python — This repo is tested with 3.6

  • Install PyTorch version >= 1.0.0, torchvision >= 0.2.1

  • Download Inception-v3 pretrained model from and move it to ./PRETRAIN/

Super-Pixel Generation

python -u

K: number of super pixels

M: compactness factor

We resize the image to 299×299 as pre-process and 400 images example can be found at here

Adversarial Attack

run on ImageNet dataset for white-box attack succes rate and robustness toward resize(2x)

python -u


View Github