MOCOGAN: training isn't runing

universome / stylegan-v

[CVPR 2022] StyleGAN-V: A Continuous Video Generator with the Price, Image Quality and Perks of StyleGAN2

https://universome.github.io/stylegan-v

333 stars 36 forks source link

MOCOGAN: training isn't runing #39

Open ParthaEth opened 8 months ago

ParthaEth commented 8 months ago

The generator seems to be barebone STG2 generator. And then it only generates a batch of images, which when fed to the video discriminator fails because of dimensionality issues. I don't see where you generate frames auto regressively.