Question About the Mechanism of Expression-Driven Facial Pose in "Expression-Driven"

[cavalier501]

Thanks much for sharing such excellent work. However, I still have some questions about how expression-driven facial pose is achieved in "Expression-Driven."

In Section 7.3.2 of the paper, the adapter layer output is element-wise multiplied with Mface to ensure that the effect of z_exp is focused solely on the facial region. However, how this preserves the expression signal injection's influence on facial pose and position.

For example, in Figure 25, the face pose and position changes after being driven, suggesting z_exp impacts both expression and facial pose.

Looking forward to your reply. Thanks!