Skip to content

Starrylay/awesome-HUST-CS-CV

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 

Repository files navigation

awesome-HUST-CS-CV

  • 华中科技大学计算机视觉实验,基于卷积神经网络的手写数字识别(利用了2015年的ILSVRC竞赛的冠军模型ResNet “残差学习” 算法),和快速梯度下降法(FGSM)的攻击以及基于对抗训练的防御。
  • 手写数字识别任务接近sota 20轮迭代下模型预测准确率可达0.996。
  • 扰动幅度为 50/255 时,非定向攻击ASR可达约 50%, 定向攻击ASR 可达约 25%.
  • STAR please! 效果展示:

定向攻击:

1 to2 2 to3 3 to4 4 to5 5 to6 6 to7 7 to8 8 to9 9 to0

About

华中科技大学计算机视觉实验,基于卷积神经网络的手写数字识别,和快速梯度下降法(FGSM)的攻击以及基于对抗训练的防御。

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors