Evaluate on unseen identities during inference

[prk9009]

Can the model generate FLAME parameters for new identities during inference that were not included in its training dataset? If not, what strategies can be employed to fine-tune the model to make it work on these unseen identities?

[SichunWU]

Hi, we do not include any identities during training, which means the face shape is unchanged. We provide code for transfer animation for other identities, please refer to the post_process folder.