CODEWITHSUNDEEP
pytorchgenerative-adversarial-network
innuendo
innuendo

Reputation: 373

GANs on color images

Most (PyTorch) open source GANs work on MNIST dataset, i.e. gray level image.

Can I use a GAN on each channel of a color image, then combine the result?

Upvotes: 1

Views: 875

Answers (1)

Jatentaki
Jatentaki

Reputation: 13113

You can just have your generator and discriminator generate and classify 3-channel images - speaking in terms of implementation, make them work on B x 3 x H x W tensors instead of B x 1 x H x W, as they do for MNIST.

You can't just use your GAN on each channel separately and concatenate at the end, because you would have no way to ensure that each channel corresponds to the same image. Say you're generating celebrity faces by first generating red channel, then green and finally blue. How would you make sure that you don't get a female sample for the red channel and a male for the green?

Upvotes: 2

Related Questions