Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formats

TL;DR

This paper introduces a geometry-aware autoencoder that effectively combines multiple 3D human pose datasets with different skeleton formats, improving pose estimation accuracy across diverse datasets.

Contribution

The paper proposes a novel affine-combining autoencoder that reduces dimensionality and enhances information sharing among skeleton formats, enabling scalable multi-dataset training.

Findings

Outperforms prior methods on multiple benchmarks.

Successfully scales to 28 datasets with different skeleton formats.

Achieves state-of-the-art results on 3DPW dataset.

Abstract

Deep learning-based 3D human pose estimation performs best when trained on large amounts of labeled data, making combined learning from many datasets an important research direction. One obstacle to this endeavor are the different skeleton formats provided by different datasets, i.e., they do not label the same set of anatomical landmarks. There is little prior research on how to best supervise one model with such discrepant labels. We show that simply using separate output heads for different skeletons results in inconsistent depth estimates and insufficient information sharing across skeletons. As a remedy, we propose a novel affine-combining autoencoder (ACAE) method to perform dimensionality reduction on the number of landmarks. The discovered latent 3D points capture the redundancy among skeletons, enabling enhanced information sharing when used for consistency regularization. Our approach scales to an extreme multi-dataset regime, where we use 28 3D human pose datasets to supervise one model, which outperforms prior work on a range of benchmarks, including the challenging 3D Poses in the Wild (3DPW) dataset. Our code and models are available for research purposes.