Hi, I have seen great results finetuned from stylegan2 hhfq checkpoint on my datasets which contains about 200 cartoon images. Among all the configs, I see that config-f shows the best result. However, the code shows that there is no svd operation in the generator or the discriminator graph. I can't see any differences between the config-f in this repo and in official stylegan2 repo but the official stylegan2 repo can't produce the same result.
I wonder what will config-f in this repo do that contributes to better results over stylegan2 repo...
Hi, I have seen great results finetuned from stylegan2 hhfq checkpoint on my datasets which contains about 200 cartoon images. Among all the configs, I see that config-f shows the best result. However, the code shows that there is no svd operation in the generator or the discriminator graph. I can't see any differences between the config-f in this repo and in official stylegan2 repo but the official stylegan2 repo can't produce the same result.
I wonder what will config-f in this repo do that contributes to better results over stylegan2 repo...
Thanks!!