Questions about the batch size 4 model about stargan-v2 HOT 8 CLOSED

@ypw1996 For stable training, you should use batch_size >= 8. We also observed the worse result when using batch_size=4.

from stargan-v2.

from stargan-v2.

@ypw1996 Thanks a lot for your interest!

from stargan-v2.

Hopefully there will be more progress later, such as more precise style control proposed by style GAN.

from stargan-v2.

Hopefully there will be more progress later, such as more precise style control proposed by style GAN.

what do you mean?

from stargan-v2.

Due to the memory limitation, I set batch_size=4, max_iter=200000, and also half the learning rate, can I get the same performance with model trained with batch_size=8?

from stargan-v2.

@ypw1996 For stable training, you should use batch_size >= 8. We also observed the worse result when using batch_size=4.

so, if I set the batch_size larger than 8 (as 32)， can get better results?

from stargan-v2.

Hopefully there will be more progress later, such as more precise style control proposed by style GAN.

what do you mean?

like this : https://github.com/woctezuma/stylegan2-projecting-images#3-expression-transfer

from stargan-v2.