psyai-net / EmoTalk_release

This is the official source for our ICCV 2023 paper "EmoTalk: Speech-Driven Emotional Disentanglement for 3D Face Animation"
Other
346 stars 33 forks source link

Paper details #15

Open DinoMan opened 1 year ago

DinoMan commented 1 year ago

Hello, I have been reading your paper and there is one detail that I do not understand. From my understanding your dataset is made up of HDTF and RAVDESS. The model in the paper mentions that the identity one-hot encoding is 24 dimensional. Do these 24 identities correspond to the actors in RAVDESS? If so how are the HDTF identities encoded. Also how does the cross reconstruction loss work with the HDTF dataset since there are no emotions and similar content in these sequences in order to apply this loss term

Zrduan724 commented 9 months ago

I have same question with you.

wenq013 commented 9 months ago

same question

hrWong commented 6 months ago

+1