natsu6767 / infogan-pytorch Goto Github PK

Hi I am currently also doing an implementation of InfoGAN and one strange bit I have observed that seems to be common across both my own and your results is that the outcome does not appear exactly as expected.

Specifically in my generated outputs, I attempted to force the ordering of numbers to go from 0 to 9 with all 0s at the top row and 9s at the bottom row, in that order. However I seem to be getting different sets of numbers with my control vector and it is consistent and systematic.

I notice your generated outputs also do not follow the sequential order, is that because your results are randomly generated per row or are you also experiencing the same technical errors?

how to run this ((After training the network to experiment with the latent code for the MNIST dataset run mnist_generate.py:

python3 mnist_generate.py --load_path /path/to/pth/checkpoint))

Hi! I want to know how to set the parameter to generate the number that i want? Is it changing the c1(the discrete latent code)? It it's correct and how to set the c1? Thanks!

Hi, thanks for your great implementation.
I run the model on CelebA dataset. However, when I test the model, I find if the categorical codes are fixed, different noise vectors introduce the same results.
How do you get the CelebA results shown in the GitHub page?

Error:

My solution:

Thank you for your project! I have a question.
I am a beginner in biology. Don't have much experience in programming.
I would like to know how to modify the latent code c(dis and con), so I could control and see the continuous features.

I know it looks like a simple problem, but I reported this error in line 108 of the document train.py. The solution on the network is to set index to long or Byte tensor. so I try to convert idx from numpy to tensor and set it to device. idx = torch.from_numpy(idx).long().to(device)But the mistake still happened. Then I found it at https://swift.ctolib.com/article/comments/96546 when np.array is assuming int32 for the indexes, but torch wants int64. I tried this method, but the problem still arose.