AvatarReX: Real-time Expressive Full-body Avatars

We present AvatarReX, a new method for learning NeRF-based full-body avatars from video data. The learnt avatar not only provides expressive control of the body, hands and face together, but also supports real-time animation rendering. To this end, we propose compositional representation, where are separately modeled in way that structural prior parametric mesh templates is properly utilized without compromising representation flexibility. Furthermore, disentangle geometry appearance each part. With these technical designs, dedicated deferred rendering pipeline, which can be executed at framerate to synthesize high-quality free-view images. disentanglement allows us design two-pass training strategy combines volume surface network training. In way, patch-level supervision applied force learn sharp details on basis estimation. Overall, our enables automatic construction with capability, generate photo-realistic images dynamic novel body motions facial expressions.