search

POSTECH-CVLab / PyTorch-StudioGAN

StudioGAN is a Pytorch library providing implementations of representative Generative Adversarial Networks (GANs) for conditional/unconditional image generation.
https://github.com/MINGUKKANG
Other
3.43k stars 345 forks source link

step_count in the generator loop #65

Closed KK666-AI closed 3 years ago

KK666-AI commented 3 years ago

Dear author,

I notice that the step_count += 1 is in the generator training loop (see line ). Does this step count is a standard usage in the research field of GAN.

Specifically, if comparing different methods over the steps, the x-axis is the the number of training steps of generator?

mingukkang commented 3 years ago

Hi,

It depends on the paper.

In the case of "Consistency regularization for Generative Adversarial Networks", they have written that "We stop training after 200k generator update steps for CIFAR10 and 100k steps for CelebA".

Thank you.

Sincerely,

KK666-AI commented 3 years ago

ok，i see, thanks.