pytorch-nips2017-attack-example icon indicating copy to clipboard operation
pytorch-nips2017-attack-example copied to clipboard

Published 20 hours ago • verified publisher icon rwightman

Num failures = 0 but still images are not adversarial

Open Puneet2000 opened this issue 6 years ago • 0 comments

I m running the CW2 attack on a model. The logs show that num_failures =0 but when I test the generated adversarial image the model is somewhat robust. Is it possible? If yes then what num failures and num success mean?

Puneet2000 avatar Aug 20 '19 08:08 Puneet2000