Conditional GAN concatenation of real image and label

vision
1

I am concatenating real images with labels and the output shape is [128, 11,28, 28]. How can I visualize this image?

encoded_labels = F.one_hot(reallabels, num_classes)
encoded_labels = encoded_labels [:, :, None, None]
encoded_labels = encoded_labels .repeat(1, 1, mnist_shape[1], mnist_shape[2])

Here the encoded_labels size is torch.Size([128, 10, 28, 28])

Now I want to concatenate it with images
cat_real_labels = torch.cat((realImages.float(), encoded_labels .float()),1)
output: torch.Size([128, 11, 28, 28])

How can I visualize this image?

2

You cannot visualize it directly, as the tensor doesn’t have a valid image shape.
However, you could visualize the 11 channels separately of all 128 samples (so 1408 grayscale images).

3

Can you just give me a hint of code? How can I visualize it separately for all 128 samples?

4

Something like this approach would work. The difference would be that you would have to flatten the first two dimensions in case you want to visualize all 11*128 grayscale images.

1 Like