Skip to content

some Textual and Image Adversarial Attack Practice(2024羊城杯AI)

Notifications You must be signed in to change notification settings

EthanC20/AdvAttackPractice

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 

Repository files navigation

AdvAttackPractice

比赛参考2024羊城杯粤港澳大湾区网络安全大赛本科组2024-YCB-Undergraduate

参赛队伍:Del0n1x;ranking:23/500(竞争太激烈力,打麻了😭)

更新:最后居然第十七名进决赛了wow😭,真是太不容易了

队伍writeup已公开访问,欢迎师傅们参考交流

NLP_Model_Attack

题目分值:已答出45次,初始分值500.0,当前分值483.2,解出分值482.43 题目难度:中等

预训练三分类模型distilBert,攻击策略为经典PWWSRen2019

使用TextAttack工具箱,详细教程和代码示例参考TextAttack Documentation — TextAttack 0.3.10 documentation,各类API文档使用方法写得很好,writeup也可以写得很简洁。

Targeted_Image_adv_attacks

题目分值:已答出10次,初始分值500.0,当前分值499.3,解出分值499.13 题目难度:困难

预训练模型为densenet121_catdogfox_classify,攻击策略为PGD,各项扰动参数为eps=4/255, alpha=0.5/255, steps=12, random_start=True

使用torchattacks工具箱,好用爱用,

重写数据加载方法,PGD CW策略参数调试,150张图片改写批量运行,三类图片分别攻击,运气也很好PGD策略能本地过测,最后提交多一点点过关。

结果通过Nvidia RTX2080TI 本地计算Average SSIM: 0.9547 Average Attack Success Rate: 0.9200

经过优化能在17s左右生成150张对抗样本且满足上述数据效果,代码是经过claude-3.5-sonnet简化的版本,环境参考AdvAttack requirements

About

some Textual and Image Adversarial Attack Practice(2024羊城杯AI)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published