Facial expression generation

End-to-end frame-level facial expression generation models.

We have the following options:

• ASAP: Augmented Self-Attention Pruning, considering facial expression context from both agent and user
• MODIFF: diffusion based adaptive facial expression generation for motivational interview (currently only works with MI Counselour DA)