DarkGeekMS/portrait-synthesis-from-speech

Explore pretrained embedding to latent vector mapping

Closed this issue · 1 comments

Use a large pretrained sentence embedding model (BERT ...,etc) to embed the text description and then train a fully connected network to map this embedding to StyleGAN latent vectors.
This approach can improve this work.

This experiment is not important anymore. Moreover, the code for the experiment exists in the mentioned repo.