GreedyFool
GreedyFool copied to clipboard
GreedyFool (NIPS2020) official code for saprse adversarial attack
GreedyFool
Implemention of GreedyFool: Distortion-Aware Sparse Adversarial Attack (NIPS2020)
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/
Adversarial Attack
run on ImageNet dataset
bash nips_gd.sh for white-box attack
bash nips_black_gd.sh for black-box attack
paramaters
--dataroot input image data with png format
--name name of the current test
--phase phase of the current test
--max_epsilon max of the adversarial perturbation [0,255]
--iter max iteration number for attack
--confidence confidence factor for control the black-box strength (do not work for the white-box setting)