Hi-LASSIE: High-Fidelity Articulated Shape and Skeleton Discovery from Sparse Image Ensemble

Instance parts Textured 3D shape or skeleton template Figure 1. Hi-LASSIE overview and sample reconstructions. Given 20-30 images of an articulated animal class, we first discover a generic 3D skeleton, then jointly optimize the camera viewpoints, skeleton articulations, as well as shared and instance-specific neural part shapes. Hi-LASSIE is able to produce high-fidelity shapes and texture without any pre-defined shape model or 3D skeleton annotations. The part-based representation also allows applications like animation and motion re-targeting.