Skip to content
Code for ICML 2019 paper "Simple Black-box Adversarial Attacks"
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
LICENSE Initial commit May 6, 2019
README.md
run_simba.py updated readme May 6, 2019
simba_single.py added 20-line implementation Jun 8, 2019
utils.py first commit May 6, 2019

README.md

This repository contains code for the ICML 2019 paper:

Chuan Guo, Jacob R. Gardner, Yurong You, Andrew G. Wilson, Kilian Q. Weinberger. Simple Black-box Adversarial Attacks.

Our code uses PyTorch (pytorch >= 0.4.1, torchvision >= 0.2.1) with CUDA 9.0 and Python 3.5. The script run_simba.py contains code to run SimBA and SimBA-DCT with various options.

To run SimBA (pixel attack):

python run_simba.py --data_root <imagenet_root> --num_iters 10000 --pixel_attack  --freq_dims 224

To run SimBA-DCT (low frequency attack):

python run_simba.py --data_root <imagenet_root> --num_iters 10000 --freq_dims 28 --order strided --stride 7

For targeted attack, add flag --targeted and change --num_iters to 30000.

You can’t perform that action at this time.