chenyuntc / pytorch-gan Goto Github PK

Good work! I have also learned GAN based on pytorch. The Cifar10 datasets has ten classes. Before using GAN, i trained Cifar10 using the following code:
for i, data in enumerate(trainloader, 0):
inputs, labels = data
inputs, labels = Variable(inputs.cuda()), Variable(labels.cuda())
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, labels)
the labels ranges from 0 to 9. and test the network, we can using these code:
for i, data in enumerate(testloader, 0):
images, labels = data
images = Variable(images.cuda())
labels = labels.cuda()
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
predicted = predicted.cuda()
correct += (predicted == labels).sum()
However, With using the GAN network trained for Cifar10, the loss = criterion(outputs, labels) has changed to criterion(outputs, real_data) or criterion(outputs, fake_data) , the ten classifications has become two classifications. So how can i correctly predicted an image belonging to which class using the Gan network?
Thanks a lot.

Hello, I would like to ask whether the results are normal when you train with mnist data set?I tried training with this dataset, and it was normal until about a dozen epochs, but then all of a sudden there was a modal collapse, and the generator produced a very strange picture.

the 'xrange' is in 'for epoch in xrange(opt.max_epoch):' is wrong.

Hi, thank you for making easy notebook for WGAN. I am trying to use your WGAN.

However, I got this error message

RuntimeError: Mismatch in shape: grad_output[0] has a shape of torch.Size([1]) and output[0] has a shape of torch.Size([32, 1, 1, 1]).

I just cloned your code, and run.

Do you know how can I fix this error?