Closed echo1993in closed 1 year ago
We also find it is hard to train StyleGAN/DualStyleGAN with a small batch size. Here is the related discussion. I think you need to tune the hyper-parameters to prevent mode collapse.
https://github.com/williamyang1991/DualStyleGAN/issues/19#issuecomment-1109899462
In finetune_dualstylegan training ,I trained with a single GPU and pre-trained models.I followed the instructions,but it was not well trained, and even the image collapsed.Is this because I am a single GPU and batch size is 1?