DCGAN : How to improve Generator and Discriminator Loss during training?

I trained a DCGAN from Pytorch tutorial that I changed a little to generate 1024 px output but since I did that Generator and Discriminator Loss during training isn’t good as you can see it here :

Do you have some advises to improve it ?

Thanks for your help ! :slight_smile:

I’m not sure if it’s that easy so scale up the resolution of the generated images.
E.g. in Progressive Growing of GANs for Improved Quality, Stability, and Variation the authors used “low-res” layers and stacked more layers on top of them after pretraining to increase the resolution.

I doubt an output of 1024x1024 will work out of the box using the DCGAN.

Now I doubt it too, thank you for your answer :slight_smile: