3D Multi-bodies: Fitting Sets of Plausible 3D Human Models to Ambiguous Image Data

Part of Advances in Neural Information Processing Systems 33 (NeurIPS 2020)

AuthorFeedback Bibtex MetaReview Paper Review Supplemental


Benjamin Biggs, David Novotny, Sebastien Ehrhardt, Hanbyul Joo, Ben Graham, Andrea Vedaldi


We consider the problem of obtaining dense 3D reconstructions of deformable objects from single and partially occluded views. In such cases, the visual evidence is usually insufficient to identify a 3D reconstruction uniquely, so we aim at recovering several plausible reconstructions compatible with the input data. We suggest that ambiguities can be modeled more effectively by parametrizing the possible body shapes and poses via a suitable 3D model, such as SMPL for humans. We propose to learn a multi-hypothesis neural network regressor using a best-of-M loss, where each of the M hypotheses is constrained to lie on a manifold of plausible human poses by means of a generative model. We show that our method outperforms alternative approaches in ambiguous pose recovery on standard benchmarks for 3D humans, and in heavily occluded versions of these benchmarks.