search

mahyarnajibi / FreeAdversarialTraining

PyTorch Implementation of Adversarial Training for Free!
Other
239 stars 42 forks source link

Well, is there anyone run it with cifar100? #8

Open NZS-YNU opened 4 years ago

NZS-YNU commented 4 years ago

this question is about train_fgsm.py I modified function "get_loaders" and net's num_classes to let program can run with cifar100. But pgd acc is only 3%. And it's normal with cifar10.

The modified part is as follows： (1)net image

(2) get_loaders image

Has anyone encountered the same problem as me?

NZS-YNU commented 4 years ago

I also set the epochs= 50

not-a-genius commented 4 years ago

I am currently using CIFAR-10 instead of 100, did you ever use it?

tsunghan-wu commented 2 years ago

Hi @NZS-YNU @not-a-genius,

Did you successfully reproduce the result of CIFAR-100?

I still cannot make it and would ask if you could provide me some hints.

Thanks!