MPL: Lifting 3D Human Pose from Multi-view 2D Poses

TL;DR

This paper introduces MPL, a transformer-based framework that lifts 3D human poses from multi-view 2D poses, effectively addressing occlusion issues and improving accuracy over triangulation methods.

Contribution

It proposes a novel transformer-based approach trained on synthetic data to improve 3D pose estimation from multi-view 2D poses, enhancing generalization to real-world scenarios.

Findings

Achieves up to 45% reduction in MPJPE errors compared to triangulation.

Effectively combines 2D pose estimation with 2D-to-3D lifting.

Demonstrates robustness in multi-view 3D human pose estimation.

Abstract

Estimating 3D human poses from 2D images is challenging due to occlusions and projective acquisition. Learning-based approaches have been largely studied to address this challenge, both in single and multi-view setups. These solutions however fail to generalize to real-world cases due to the lack of (multi-view) 'in-the-wild' images paired with 3D poses for training. For this reason, we propose combining 2D pose estimation, for which large and rich training datasets exist, and 2D-to-3D pose lifting, using a transformer-based network that can be trained from synthetic 2D-3D pose pairs. Our experiments demonstrate decreases up to 45% in MPJPE errors compared to the 3D pose obtained by triangulating the 2D poses. The framework's source code is available at https://github.com/aghasemzadeh/OpenMPL .