search

locuslab / fast_adversarial

[ICLR 2020] A repository for extremely fast adversarial training using FGSM
434 stars 92 forks source link

reproduce problem of imagenet on default set #19

Closed liuxingbin closed 3 years ago

liuxingbin commented 3 years ago
In table 4. imagenet adversairal robustness result: method epsilon pgd+1restart
FGSM 2/255 43.46%
FGMS 4/255 30.28%

Is this top5 accuracy or top1 accuracy?

riceric22 commented 3 years ago

This is Top1 accuracy