-
Notifications
You must be signed in to change notification settings - Fork 23
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Can you share the training files? #2
Comments
When testing the classifier with the original test set, are you using normalization? |
Yes, data normalization(mean=[0.4914, 0.4822, 0.4465], std=[0.2023, 0.1994, 0.2010]) is used and batch normalization is turned off because of model.eval(). Now the test accuracy is only 81.5%. The resnet50 model I used is modified for cifar10: https://github.com/bearpaw/pytorch-classification/blob/master/models/cifar/resnet.py. I think there are some mistakes. Could you please point it out? Thanks a lot! |
How did you get the Robust accuracy? Is it the same as the paper? |
I only tried to reproduce the standard accuracy but what I got was not so good. |
First I really appreciate your work and it is really helpful to understand the essence of adversarial examples.
I met some problems when I try to re-implemented the standard testing results of the model trained with Non-Robust Cifar10 Dataset. I trained the Non-Robust Cifar10 Dataset as you said in Appendix C.2 but I only got a bad test accuracy(about 64% which is not 88%).
Could you please share your training code? Thanks in advance and I can't help saying that this paper is extremely useful!
The text was updated successfully, but these errors were encountered: