search

Harry24k / adversarial-attacks-pytorch

PyTorch implementation of adversarial attacks [torchattacks]
https://adversarial-attacks-pytorch.readthedocs.io/en/latest/index.html
MIT License
1.92k stars 353 forks source link

[QUESTION]BIM and PGD have the same output #185

Open pleaseWind opened 6 months ago

pleaseWind commented 6 months ago

❔ Any questions

I use your bim and pgd code to test my cnn model in MNIST, but get the same accuracy. Is it true?

rikonaka commented 6 months ago

Hi @pleaseWind , can you provide your code?

pleaseWind commented 6 months ago

I have uploaded the file to my Repository. I find the reason may be the PGD's parameter"random_start". Although I modify code, the PGD's accuracy is lower than BIM's. This is different from the literature.

rikonaka commented 6 months ago

Hi @pleaseWind , can you provide the link to your code?

pleaseWind commented 6 months ago

https://github.com/pleaseWind/adversarial-attacks-pytorch

rikonaka commented 6 months ago

Hi @pleaseWind , MNIST as a dataset is too simple at the moment, so the success rate of attacking such a dataset is basically 100%, and if you want to get a different result, you can test on CIFAR-10 and larger datasets.

pleaseWind commented 6 months ago

OK, I will try. Thank for your help.