Conceptdisentanglement
Disentanglement
A representation learning approach that separates expression semantics from view-related information, so that the encoding channel carries only expression semantics while view information is carried by an explicit camera-parameter channel.
Timeline (last 90 days)
Oct 19
Disentanglement can eliminate the view bias of learned expression encodings, enabling the encoding to serve as a driving signal for multi-view consistent avatars.
Unverified50%