znxlwm / pytorch-mnist-celeba-gan-dcgan Goto Github PK

Sir, as u said the avg time for an epoch is around 180s, while on my server, it shows:
[1/20] - ptime: 372.38, loss_d: 0.597, loss_g: 5.759

My environment is:
ubuntu 16.04+cuda8.0+cudnn 6+ pytorch 0.2 +Titan XP

I also set the worker_num for train data loader to 2, so it shouldn't be a problem of IO.
Do u have any idea of what's going wrong , Sir?

I check your DCGAN implementation details.Why is the size of generated mnist image is (64,64) rather than (28,28)?

thanks.

Hi, dear author:
I found your figure in readme.md looks pretty good:

I wonder what tool u used to make it? .

Dear Mr:
I'm interested in your work and want to know how to get the Croped image.

Do u know how to request for the CelebA Identity List?
Will be glad if u have it!

Hello, your code is very good, but I have some problems that I can't solve.When I replace

transform = transforms.Compose([
transforms.Scale(img_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5], std=[0.5])
])
and
D_train_loss.item()

And then you get these two warnings

UserWarning: volatile was removed and now has no effect. Use with torch.no_grad(): instead.
fixed_z_ = Variable(fixed_z_.cuda(),volatile=True)
UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.
"please use transforms.Resize instead.")

and after that the program keeps running, but it doesn't do any calculations like output losses and so forth.I would be most grateful if you could answer me

i used python3.5 and pytorch0.4 which can successfully train the CelebA.but when i turned the parameter “IsCrop=True” ,there is another error “Runtime Error:sizes must be non-negative ”
Do you know how can I fix it?hope to get your help

for x_, _ in train_loader:
    # train discriminator D
    D.zero_grad()

    x_ = x_.view(-1, 28 * 28)

    mini_batch = x_.size()[0]

When I tried to run these lines, it showed me " output with shape [1, 28, 28] doesn't match the broadcast shape [3, 28, 28]" and I don't know why.

Could you help me to fix this bug?

And by the way, in the loop "for x_, _ in train_loader", what is the structure of x_ and ? I could not get the meaning of x and _

I am new to GAN. Thanks!

In celebA project,

ax[i, j].imshow((test_images[k].cpu().data.numpy().transpose(1, 2, 0) + 1) / 2)

is there any reason for you to make the image (value +1)/2?

Existing configuration :
Python 2.7.6
pytorch 0.1.12

Hello
I ran your code of DCGAN implementation on dataset of MNIST but the quality of the generated images were poor. I have already tried to reduce the learning rate but it didn't work and the result was a far cry from yours. I am new to GAN and feel really confused. Did you change your parameter settings or do some other adjustments? Hope to get some advice from you, thank you very much!