Learning Temporal 3D Human Pose Estimation with Pseudo-Labels

TL;DR

This paper introduces a self-supervised approach for 3D human pose estimation that leverages temporal information and multi-view triangulation to improve accuracy, achieving state-of-the-art results on standard benchmarks.

Contribution

It proposes a novel method combining temporal convolutional networks with multi-view triangulation for self-supervised 3D human pose estimation.

Findings

Achieves state-of-the-art results on Human3.6M and MPI-INF-3DHP datasets.

Effectively uses temporal and multi-view information during training.

Operates with single-view 2D inputs during inference.

Abstract

We present a simple, yet effective, approach for self-supervised 3D human pose estimation. Unlike the prior work, we explore the temporal information next to the multi-view self-supervision. During training, we rely on triangulating 2D body pose estimates of a multiple-view camera system. A temporal convolutional neural network is trained with the generated 3D ground-truth and the geometric multi-view consistency loss, imposing geometrical constraints on the predicted 3D body skeleton. During inference, our model receives a sequence of 2D body pose estimates from a single-view to predict the 3D body pose for each of them. An extensive evaluation shows that our method achieves state-of-the-art performance in the Human3.6M and MPI-INF-3DHP benchmarks. Our code and models are publicly available at \url{https://github.com/vru2020/TM_HPE/}.