Super_Pix_Adv
Offical implemention of Robust Superpixel-Guided Attentional Adversarial Attack (CVPR2020)
Setup
-
Install
python
-- This repo is tested with3.6
-
Install
PyTorch version >= 1.0.0, torchvision >= 0.2.1
-
Download Inception-v3 pretrained model from
https://download.pytorch.org/models/inception_v3_google-1a9a5a14.pth
and move it to./PRETRAIN/
Super-Pixel Generation
python -u get_slic.py
K: number of super pixels
M: compactness factor
We resize the image to 299x299 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 abim_sup.py