search

microsoft / StyleSwin

[CVPR 2022] StyleSwin: Transformer-based GAN for High-resolution Image Generation
https://arxiv.org/abs/2112.10762
MIT License
508 stars 49 forks source link

Training log of losses #16

Closed HAN-oQo closed 2 years ago

HAN-oQo commented 2 years ago

Hi, Thank you for your awesome research! I am training the model with my own dataset, i want to know the training logs of losses if it is possible. Discriminator loss seems to converge so fast (close to 0) , is it right?

Best regards, Hankyu Jang

ForeverFancy commented 2 years ago

Hi, thanks for your interest of our work. The loss of D would not converge to 0 under normal conditions. Which loss logs do you want to know, I could send them to your mail.

HAN-oQo commented 2 years ago

Thank you for your answer. Can I know the graphs of G, D losses in the tensorboard, plz?

ForeverFancy commented 2 years ago

Hi, part of the training logs of FFHQ-256 could be viewed here. Loss of G, loss of D and r1_val.

HAN-oQo commented 2 years ago

Thank you!