inference face error

[LinYi-JCY]

I trained with the data of more than 200 people, all the training data was checked, and the training data was not corrupted.

sync loss has been reduced to about 0.18, some faces can be inferred normally, but some faces will have garbled characters.

What might be the cause of this? How to fix it?

[ghost]

you need to train the generator as you can see the quality is good then apply syncnet