Header

UZH-Logo

Maintenance Infos

Learning Monocular 3D Human Pose Estimation from Multi-view Images


Rhodin, Helge; Meyer, Frédéric; Spoerri, Jörg; Müller, Erich; Constantin, Victor; Katircioglu, Isinsu; Fua, Pascal; Salzmann, Mathieu (2018). Learning Monocular 3D Human Pose Estimation from Multi-view Images. In: IEEE/CVF 2018 Conference on Computer Vision and Pattern Recognition, Salt Lake City, 18 June 2018 - 23 June 2018.

Abstract

Accurate 3D human pose estimation from single images is possible with sophisticated deep-net architectures that have been trained on very large datasets. However, this still leaves open the problem of capturing motions for which no such database exists. Manual annotation is tedious, slow, and error-prone. In this paper, we propose to replace most of the annotations by the use of multiple views, at training time only. Specifically, we train the system to predict the same pose in all views. Such a consistency constraint is necessary but not sufficient to predict accurate poses. We therefore complement it with a supervised loss aiming to predict the correct pose in a small set of labeled images, and with a regularization term that penalizes drift from initial predictions. Furthermore, we propose a method to estimate camera pose jointly with human pose, which lets us utilize multiview footage where calibration is difficult, e.g., for pan-tilt or moving handheld cameras. We demonstrate the effectiveness of our approach on established benchmarks, as well as on a new Ski dataset with rotating cameras and expert ski motion, for which annotations are truly hard to obtain.

Abstract

Accurate 3D human pose estimation from single images is possible with sophisticated deep-net architectures that have been trained on very large datasets. However, this still leaves open the problem of capturing motions for which no such database exists. Manual annotation is tedious, slow, and error-prone. In this paper, we propose to replace most of the annotations by the use of multiple views, at training time only. Specifically, we train the system to predict the same pose in all views. Such a consistency constraint is necessary but not sufficient to predict accurate poses. We therefore complement it with a supervised loss aiming to predict the correct pose in a small set of labeled images, and with a regularization term that penalizes drift from initial predictions. Furthermore, we propose a method to estimate camera pose jointly with human pose, which lets us utilize multiview footage where calibration is difficult, e.g., for pan-tilt or moving handheld cameras. We demonstrate the effectiveness of our approach on established benchmarks, as well as on a new Ski dataset with rotating cameras and expert ski motion, for which annotations are truly hard to obtain.

Statistics

Citations

Dimensions.ai Metrics
1 citation in Web of Science®
6 citations in Scopus®
Google Scholar™

Altmetrics

Downloads

0 downloads since deposited on 15 Feb 2019
0 downloads since 12 months

Additional indexing

Item Type:Conference or Workshop Item (Speech), not_refereed, further contribution
Communities & Collections:04 Faculty of Medicine > Balgrist University Hospital, Swiss Spinal Cord Injury Center
Dewey Decimal Classification:610 Medicine & health
Language:English
Event End Date:23 June 2018
Deposited On:15 Feb 2019 14:59
Last Modified:29 Oct 2019 08:24
Publisher:IEEE
OA Status:Closed
Publisher DOI:https://doi.org/10.1109/CVPR.2018.00880

Download

Closed Access: Download allowed only for UZH members