Thank you for your great project!
Have you tried to train model on Celeba on smaller image sizes for example on 64x64? On 256x256 i can get diverse results, but on 64x64 output is almost identical. I also tried to increase KL weight, but the diversity is poor.
Do you have any suggestions how to get diverse results on 64x64?
Thank you for your great project! Have you tried to train model on Celeba on smaller image sizes for example on 64x64? On 256x256 i can get diverse results, but on 64x64 output is almost identical. I also tried to increase KL weight, but the diversity is poor.
Do you have any suggestions how to get diverse results on 64x64?