SDIT
SDIT copied to clipboard
is inference with non-square HQ images possible?
Hi, thanks for sharing this great project. Is inference with non-square HQ images possible?
Hi, I think it is possible. To be honest, I did not try. I guess it is easy to adapt it if it fail
How can I achieve that? Should I train with CelebHQ with 1024px images?
I would like you try 512px to get insight, since I try 256px before, the results still be interesting. Or maybe you try 256px to obtain the trained model(step1), then you increase image resolution two times, and perform fine-tuning (step2). Finally you train the model based on step2 for 1024px image. I guess it has batter performance.