Curls & Whey: Boosting Black-Box Adversarial Attacks (accepted to CVPR2019)
============================================================
Introduction
------------------------------------------------------------
Image classifiers based on deep neural networks suffer from harassment caused by adversarial examples. Two defects exist in black-box iterative attacks that generate adversarial examples by incrementally adjusting the noise-adding direction for each step. On the one hand, existing iterative attacks add noises monotonically along the direction of gradient ascent, resulting in a lack of diversity and adaptability of the generated iterative trajectories. On the other hand, it is trivial to perform adversarial attack by adding excessive noises, but currently there is no refinement mechanism to squeeze redundant noises. In this work, we propose Curls & Whey black-box attack to fix the above two defects. During Curls iteration, by combining gradient ascent and descent, we ‘curl’ up iterative trajectories to integrate more diversity and transferability into adversarial examples. Curls iteration also alleviates the diminishing marginal effect in existing iterative attacks. The Whey optimization further squeezes the ‘whey’ of noises by exploiting the robustness of adversarial perturbation. Extensive experiments on Imagenet and Tiny-Imagenet demonstrate that our approach achieves impressive decrease on noise magnitude in l2 norm. Curls & Whey attack also shows promising transferability against ensemble models as well as adversarially trained models. In addition, we extend our attack to the targeted misclassification, effectively reducing the difficulty of targeted attacks under black-box condition.
Paper
------------------------------------------------------------
Yucheng Shi, Siyu Wang, Yahong Han. "Curls & Whey: Boosting Black-Box Adversarial Attacks." CVPR 2019 (Oral).
https://arxiv.org/abs/1904.01160
Reference
------------------------------------------------------------
If you find this useful in your work, please consider citing the following reference:
```
@inproceedings{CurlsWhey2019CVPR,
title = {Curls & Whey: Boosting Black-Box Adversarial Attacks},
author = {Shi, Yucheng and Wang, Siyu and Han, Yahong},
booktitle = {Computer Vision and Pattern Recognition (CVPR), 2019},
year = {2019}
}
```
Datasets
------------------------------------------------------------
The datasets used in the paper are available at the following links:
* [Imagenet](http://image-net.org/index)
* [Tiny Imagenet](https://tiny-imagenet.herokuapp.com/)
Environment
------------------------------------------------------------
The code is developed using python 3.5 and pytorch 0.4.1 on Ubuntu 16.04. NVIDIA GPUs are needed. The code is developed and tested using 1 GeForce GTX TITAN X GPU cards.
Usage
------------------------------------------------------------
./bmodels/inceptionv3/inceptionv3.pt    (https://pan.baidu.com/s/1_j7gVcGcWaobgJi7K11e6A)      code: tcax <Br/>
./fmodels/resnet/resnet101.pt           (https://pan.baidu.com/s/19kQBVwhtZw4mgHuarwFQjQ)   code: z2w9 <Br/>
./temp.zip                            (https://pan.baidu.com/s/1CMvpGyGKwFpKV1lPhr4FUA)    code: qk6y <Br/>
<Br/>
unzip temp.zip <Br/>
pip --no-cache-dir install -r requirements.txt <Br/>
<Br/>
python untargeted_attack.py <Br/>
python targeted_attack.py
Examples
------------------------------------------------------------
<br/>
“Curls&WheyBoostingBlack-BoxAdversarialAttacks”在pytorch中的实施.zip
版权申诉
97 浏览量
2023-03-31
22:35:55
上传
评论
收藏 6.58MB ZIP 举报
“Curls&WheyBoostingBlack-BoxAdversarialAttacks”在pytorch中的实施.zip （220个子文件） 
.gitignore 142B LICENSE 11KB README.md 4KB 094.npy 12KB 089.npy 12KB 098.npy 12KB 005.npy 12KB 183.npy 12KB 004.npy 12KB 180.npy 12KB 031.npy 12KB 003.npy 12KB 177.npy 12KB 182.npy 12KB 149.npy 12KB 013.npy 12KB 171.npy 12KB 084.npy 12KB 022.npy 12KB 132.npy 12KB 193.npy 12KB 103.npy 12KB 006.npy 12KB 012.npy 12KB 136.npy 12KB 083.npy 12KB 130.npy 12KB 009.npy 12KB 050.npy 12KB 154.npy 12KB 113.npy 12KB 191.npy 12KB 076.npy 12KB 025.npy 12KB 115.npy 12KB 088.npy 12KB 075.npy 12KB 186.npy 12KB 001.npy 12KB 162.npy 12KB 028.npy 12KB 152.npy 12KB 080.npy 12KB 019.npy 12KB 074.npy 12KB 184.npy 12KB 091.npy 12KB 128.npy 12KB 118.npy 12KB 163.npy 12KB 055.npy 12KB 133.npy 12KB 042.npy 12KB 097.npy 12KB 195.npy 12KB 112.npy 12KB 141.npy 12KB 020.npy 12KB 116.npy 12KB 192.npy 12KB 197.npy 12KB 082.npy 12KB 035.npy 12KB 053.npy 12KB 175.npy 12KB 150.npy 12KB 190.npy 12KB 189.npy 12KB 140.npy 12KB 134.npy 12KB 056.npy 12KB 173.npy 12KB 198.npy 12KB 064.npy 12KB 185.npy 12KB 046.npy 12KB 086.npy 12KB 110.npy 12KB 023.npy 12KB 111.npy 12KB 129.npy 12KB 014.npy 12KB 166.npy 12KB 099.npy 12KB 167.npy 12KB 048.npy 12KB 065.npy 12KB 124.npy 12KB 008.npy 12KB 095.npy 12KB 049.npy 12KB 066.npy 12KB 093.npy 12KB 179.npy 12KB 007.npy 12KB 079.npy 12KB 044.npy 12KB 072.npy 12KB 114.npy 12KB 135.npy 12KB共 220 条
- 1
- 2
- 3
资源评论
快撑死的鱼
- 粉丝: 1w+
- 资源: 9154
