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

Source: [3D Multi-bodies: Fitting Sets of Plausible 3D Human Models to Ambiguous Image Data](<https://www.sota2.com/research/paper/68fa68aa1e209131cea1d95a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T03:21:38.770Z

Authors: Benjamin Biggs, S\\'ebastien Ehrhadt, Hanbyul Joo, Benjamin Graham, Andrea Vedaldi, David Novotny

Published: 2020-11-02T13:55:31.000Z

## Abstract

We consider the problem of obtaining dense 3D reconstructions of humans 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 modelled 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Human Mesh Recovery on 3DPW standard (test)](<https://www.sota2.com/research/sota/human-mesh-recovery-on-3dpw-standard-test.md>): Human Mesh Recovery; 3DPW standard (test); RE: 55.6.

- [Human Mesh Recovery on Human3.6M Protocol #2 (S9, S11 test)](<https://www.sota2.com/research/sota/human-mesh-recovery-on-human3-6m-protocol-sharp-2-s9-s11-test.md>): Human Mesh Recovery; H36M Protocol #2 S9, S11 (test); Reconstruction Error (RE): 41.6.

- [Human Mesh Recovery on Human3.6M (test)](<https://www.sota2.com/research/sota/human-mesh-recovery-on-human3-6m-test.md>): Human Mesh Recovery; Ambiguous H36M (test); MPJPE: 90.

- [3D Human Mesh Recovery on Human3.6M (test)](<https://www.sota2.com/research/sota/3d-human-mesh-recovery-on-human3-6m-test.md>): 3D Human Mesh Recovery; Human3.6M (test); PA-MPJPE: 42.2.

- [3D Human Mesh Recovery on 3DPW (test)](<https://www.sota2.com/research/sota/3d-human-mesh-recovery-on-3dpw-test.md>): 3D Human Mesh Recovery; 3DPW (test); PA-MPJPE: 55.6.

- [3D Human Pose Estimation on 3DPW (test)](<https://www.sota2.com/research/sota/3d-human-pose-estimation-on-3dpw-test.md>): 3D Human Pose Estimation; 3DPW (test); PA-MPJPE: 59.9.

- [3D Human Pose Estimation on Human3.6M (test)](<https://www.sota2.com/research/sota/3d-human-pose-estimation-on-human3-6m-test.md>): 3D Human Pose Estimation; Human3.6M (test); MPJPE (Average): not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2011.00980>)

[PDF](<https://arxiv.org/pdf/2011.00980.pdf>)