Zenan Ling comments

Results 7 comments of


                                            Zenan Ling

Is the CNN version code trained well?

> My conjecture is that the optimization step makes spectral norm larger than 1 and your code uses sigma calculated in the training phase to normalize it. It changes weight...

Is the CNN version code trained well?

@shimazing yes

Is the CNN version code trained well?

@jarrelscy Problems still exit without data parallel. Here is a toy example. import torch.nn as nn from SpectralNormGouk1 import * from torch.optim import * class toy(nn.Module): def __init__(self): super(toy, self).__init__()...

Is the CNN version code trained well?

@jarrelscy Thanks for your reply.

Is the CNN version code trained well?

@jarrelscy The test loss and accuracy seem to be normal if I use "net.train()" and "with with torch.no_grad()" during the test phase.

Is the CNN version code trained well?

@shimazing @jarrelscy did you train the classification model？The author release the code in the latest version paper but the link is 404 now.

Is the CNN version code trained well?

my classification net doesn’t work on single gpu the loss explodes