G. Hod et al., "Neural Approaches for 3D Pose Estimation from 3D Data", Proc. of 3DBODY.TECH 2023 - 14th Int. Conf. and Exh. on 3D Body Scanning and Processing Technologies, Lugano, Switzerland, 17-18 Oct. 2023, #42, https://doi.org/10.15221/23.42.

Title:

Neural Approaches for 3D Pose Estimation from 3D Data

Authors:

Gali HOD 1, Tal BARAMI 1,2, Michael KOLOMENKIN 1

1 Playtika LTD, Herzliya, Israel;
2 Department of Computer Science, Ben-Gurion University of the Negev, Beersheba, Israel

Abstract:

Understanding human pose is a fundamental component of many forms of art, including sculpture, painting, drawing, and animation. Software that can accurately capture and represent the human pose is essential for creating realistic and expressive works of both traditional and digital art. However, to the best of our knowledge, there is currently no open-source code available for deep learning-based pose estimation from static 3D data. While there exist many "classical" pre-deep learning methods for this task, they have a significant drawback: they are not differentiable, making them difficult to incorporate into subsequent deep learning pipelines. We put special emphasis on integration with deep learning pipelines since they are the cornerstone of modern creative systems. In this work, we propose and implement two methods for human pose estimation based on neural networks. The first method leverages part segmentation to classify the body part of each point and estimates the body joints based on neighboring parts. The second method estimates joints directly from point clouds. Our code will be made available on GitHub.

Keywords:

Human 3D pose estimation, neural network, point cloud, mesh, skeleton reconstruction

Details/PDF/VIDEO:

Full paper: PDF
Proceedings: 3DBODY.TECH 2023, 17-18 Oct. 2023, Lugano, Switzerland
Paper id#: 42
DOI: 10.15221/23.42

Copyright notice

© Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Reproduction of the proceedings or any parts thereof (excluding short quotations for the use in the preparation of reviews and technical and scientific papers) may be made only after obtaining the specific approval of the publisher. The papers appearing in the proceedings reflect the author's opinions. Their inclusion in these publications does not necessary constitute endorsement by the editor or by the publisher. Authors retain all rights to individual papers.

Note: click the + on the top left of the page to open/close the menu.