MetaPose: Fast 3D Pose from Multiple Views without 3D Supervision

TL;DR

MetaPose introduces a fast, multi-view 3D human pose estimation method that requires only 2D keypoints for training, effectively handling occlusions and outperforming existing approaches on standard and in-the-wild datasets.

Contribution

The paper presents a neural model for multi-view 3D pose estimation that operates without 3D supervision and accounts for joint uncertainty, achieving high accuracy with minimal latency.

Findings

Outperforms classical bundle adjustment and weakly-supervised baselines

Works effectively on Human3.6M and Ski-Pose PTZ datasets

Requires only 2D keypoints for training

Abstract

In the era of deep learning, human pose estimation from multiple cameras with unknown calibration has received little attention to date. We show how to train a neural model to perform this task with high precision and minimal latency overhead. The proposed model takes into account joint location uncertainty due to occlusion from multiple views, and requires only 2D keypoint data for training. Our method outperforms both classical bundle adjustment and weakly-supervised monocular 3D baselines on the well-established Human3.6M dataset, as well as the more challenging in-the-wild Ski-Pose PTZ dataset.