Generalization

[leg0m4n]

Hi, thank you for the paper and open-sourcing code. From my understanding, the model works only on the dataset it was trained on and any audio/head pose/blink signal, so it can not be applied to a random video of a never-seen talking person in it, right?

Can you please share your thoughts about what can be done to make the model applicable to a never seen before video? Thank you.