perceptual-advex

perceptual-advex copied to clipboard

Have you tried to develop a better metric than LPIPS for perceptibility? I was wondering if there's room for developing a better perceptibility metric or if you've already tried hard...

futakw

It seems that the defense model is a PAT-adversarially trained model. When attacking a normally trained model (clean model), the perturbation will be very similar to l2 attack ? (especially...

Equationliu

Robust Accuracy of StAdv Model

7

Hi, according to Table 2, (AT StAdv) model achieves >50% accuracy against StAdv attacks (bound = 0.05). I used the pre-trained models provided (stadv_0.05.pt) but got a robust accuracy of...

pratyushmaini

A mistake which may lead to performance degradation

2

https://github.com/cassidylaidlaw/perceptual-advex/blob/70af18f174a269ed17368116b41e59e10cad0195/perceptual_advex/perceptual_attacks.py#L623 We should update "perturbations" after we update "live". Otherwise, some samples which already meet the requirements will still be updated through https://github.com/cassidylaidlaw/perceptual-advex/blob/70af18f174a269ed17368116b41e59e10cad0195/perceptual_advex/perceptual_attacks.py#L633

kevinbro96