search

guanjz20 / StyleSync_PyTorch

PyTorch implementation of "StyleSync: High-Fidelity Generalized and Personalized Lip Sync in Style-based Generator"
202 stars 22 forks source link

Details of training syncnet #2

Open sunwoo76 opened 1 year ago

sunwoo76 commented 1 year ago

Hello :)

I have a question about the training on your own syncnet.

As far as I know, it is difficult to converge the losses of Syncnet. So, many users have left messages in the Issues tab of the Wav2Lip GitHub repository.

Could you explain the architecture or training scheme of Syncnet of this model in more detail?

Thanks! :_)

spark-hudson commented 6 months ago

I am also interested in knowing how the syncnet is trained. Is eq (4) in the paper the same loss that was used during the training of syncnet?

Thanks in advance.

PolarRobin commented 5 months ago

Would love get some details as well :)