Animatable NeRF is the latest outcome of ZJU 3D Vision Lab. Animatable NeRF reconstructs an animatable human model from a multi-view video. They introduce neural blend weight fields to produce the deformation fields. Based on the skeleton-driven deformation, blend weight fields are used with 3D human skeletons to generate observation-to-canonical and canonical-to-observation correspondences.
NeuS leverages zero-level set of a signed distance function (SDF) to represent an object’s surface. With density distribution induced by SDF，it optimizes volume rendering and proposes a new neural implicit surface representation.
