search

Hangz-nju-cuhk / Talking-Face_PC-AVS

Code for Pose-Controllable Talking Face Generation by Implicitly Modularized Audio-Visual Representation (CVPR 2021)
Creative Commons Attribution 4.0 International
912 stars 173 forks source link

About non-id space training. #28

Open NNNNAI opened 3 years ago

NNNNAI commented 3 years ago

Thanks for sharing your work.Could you tell me how you train the non-id speace and what kind of loss did you use.Cause I find that there are seems not loss for non-id space training have been mentioned in the paper. Many thanks.

Hangz-nju-cuhk commented 3 years ago

Normally no loss is needed for this space learning. I found later that the non-identity space could be trained faster in the Face Reenactment setting, which is described in the paper "Neural head reenactment with latent pose descriptors." CVPR 2020.