Monocular 3D Human Pose Estimation via Euler Angles

ICLR 2026 Conference Submission17390 Authors

Keywords: 3D human pose estimation, Joint rotation, Euler angle, body's self-rotation discontinuities, orientation conditions

TL;DR: 3D human pose estimation using Euler angles addresses the problem of body's discontinuous rotations by pre-estimated conditions.

Abstract: Monocular 3D human pose estimation is a key problem in computer vision. Existing joint position-based methods often suffer from the issues of accurate bone length prediction and rotation ambiguities when joints are collinear. Joint rotation-based methods can avoid the bone length issue but encounter discontinuities when predicting body’s self-rotation angles, limiting their applicability. In this work, we theoretically analyze the root cause of the discontinuity and propose a conditional Euler angle-based estimation method. Our approach projects the continuous body self-rotation angle in a high-dimensional space into a two-dimensional space and divides the angle into discrete angle intervals. A classification network learns the prior information about the body’s orientations in these discrete angle intervals. Then, the orientation conditions are used as inputs to improve the prediction of the Euler angle. Experiments across diverse models, including CNNs, GCNs, and Transformers, demonstrate that our method produces continuous self-rotation prediction. It effectively resolves the discontinuity problem of Euler angles and forms a plug-and-play module for efficient model transfer.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 17390