bfarzin / pytorch_aae Goto Github PK

Hi,
Could you please point to where the trained AdvAE weights are being loaded in main.py before the features are being classified into the MNIST classes?
Thanks

Hi. Thanks for your awesome codes.

I'm studying AAE, so I try to run your codes.

But when i see Discriminator loss and Generator loss, i think shape is weird.

(I can't use tensorflow logger in your code, so I make another tensorboard. But same value is used.)

Is it okay??

Hello, in sampling a random gaussian latent vector, could you please tell me why you multiply it by 5 using

Variable(torch.randn(images.size()[0], z_red_dims) * 5.).cuda()

instead of ,

Variable(torch.randn(images.size()[0], z_red_dims) ).cuda()

Moreover, in this case we cannot ensure each dimension of latent vector to be a tensor in range (0,1). My confusion is, how can the gaussian discriminator still work without a normalized gaussian vector or latent vector as its input?

Hi, thanks for the nice work.

Can you explain about distributions and histograms in tensorboard ? How can I understand?