taki0112 / StyleGAN-Tensorflow

Simple & Intuitive Tensorflow implementation of StyleGAN (CVPR 2019 Oral)
MIT License
211 stars 59 forks source link

Spectral norm + equalized learning rate? #8

Open capilano opened 5 years ago

capilano commented 5 years ago

The original Style Gan implementation does not have spectral norm. I am not sure how spectral norm interacts with the equalized learning rate multiplier. Did you observe improvements by adding spectral norm?

rfitz123 commented 5 years ago

Also wondering how to implement spectral normalization. I'm having a lot of problems with stability.