conditionwang/ntl

This is the code of ICLR 2022 Oral paper 'Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization'.

ntl's Issues

wrong discriminator is freezed

In L468 of gan_aug_cifar_sm.py, the model that should be frozen is discriminator_aug, while the one that is actually frozen is discriminator. Obviously, the discriminator does not appear below, while discriminator_aug not only appears, but also be updated by g_loss, so I would like to know whether the "Freeze D" mentioned in the proposed algorithm is implemented correctly?

This problem seems also occur in the code of GANs corresponding to other datasets.

question

Please don't make promises casually. If you don't plan to release the source code, you can make no promises. It's been two times that you've broken your promises. Don't make excuses that you're busy.

Could you please share the checkpoints of the generator models for gan_aug?

Hi,
Thanks for your great work!
Could you please share the checkpoints of the generator models for gan_aug? And what's the value of n_epochs for training GAN for augmentation?
Thanks!