Did you update your pretrain model weight??

[rosinality]

Did you update your pretrain model weight??

[rosinality]

It is paired checkpoints. Do you mean running average of generator? It is not paired but I think you can make it using generator/discriminator checkpoints as it is running average of generator weights during training.

[Johnson-yue]

how much size of input image for this generator/discirminator checkpoint ?

[rosinality]

Input image size is 256 pixels.

[Johnson-yue]

Thank you , I will finetune g and d on my datasets .

[Johnson-yue]

@rosinality which config should I use? ** for celebA** or ** for FFHQ** with pretrained weight

[rosinality]

You should use FFHQ.

[Johnson-yue]

reply so quickly!!! thanks a lot!!

[Johnson-yue]

@rosinality the code default "max_size" is 1024, default "init_size" is 8.
If I train in my dataset that. input image size is 256^2, and I want to generate size is also 256^2
, should I change max_size to 256 and init_size to 4 ?

[rosinality]

You only need to change max_size if you need to reduce final image sizes.

[Johnson-yue]

ok, 🙏 thanks