View in EDS

NovelPoseNet: Synthesizing novel views of 2D poses for absolute and relative monocular 3D human pose estimation.

Saved in:
Bibliographic Details
Title: NovelPoseNet: Synthesizing novel views of 2D poses for absolute and relative monocular 3D human pose estimation.
Authors: Upadhyay, Avinash1 (AUTHOR) e21soep0026@bennett.edu.in, Shukla, Ankit1 (AUTHOR) v-ankit.shukla@bennett.edu.in, Sharma, Manoj1 (AUTHOR) manoj.sharma1@bennett.edu.in
Source: Pattern Recognition Letters. Apr2026, Vol. 202, p141-148. 8p.
Subjects: Triangulation, Artificial neural networks, Three-dimensional imaging
Abstract: • Novel approach: Proposed a new approach to monocular 3D pose estimation using viewpoint information as prior. • Multi-modal network: Introduced a novel ResNet-transformer network for pose synthesis from 2D pose latent information and hypothetical camera parameters. • Triangulation-based 3D Reconstruction: Used triangulation to reconstruct 3D pose from novel poses generated from a monocular image. • Absolute 3D pose estimation: Estimated absolute 3D pose from monocular images. Monocular 3D human pose estimation aims to recover the 3D joint coordinates of a person from a single 2D image, a task made inherently difficult by the loss of depth cues during image projection. To overcome the resulting depth ambiguity, we propose NovelPoseNet, a multi-modal ResNet-Transformer framework that explicitly generates novel 2D poses from unseen viewpoints conditioned on learned 2D pose features and hypothetical camera extrinsics. Unlike prior works that impose multi-view consistency as a supervision constraint, NovelPoseNet synthesizes explicit 2D pose representations from new camera perspectives, thereby enabling true multi-view triangulation from a single image. The reconstructed 3D pose is obtained by fusing these synthesized viewpoints through geometric triangulation, yielding both absolute and relative 3D joint positions. Extensive experiments on Human3.6M and MPI-INF-3DHP datasets demonstrate that NovelPoseNet achieves state-of-the-art performance, surpassing existing reprojection- and synthesis-based methods. [ABSTRACT FROM AUTHOR]
Copyright of Pattern Recognition Letters is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:• Novel approach: Proposed a new approach to monocular 3D pose estimation using viewpoint information as prior. • Multi-modal network: Introduced a novel ResNet-transformer network for pose synthesis from 2D pose latent information and hypothetical camera parameters. • Triangulation-based 3D Reconstruction: Used triangulation to reconstruct 3D pose from novel poses generated from a monocular image. • Absolute 3D pose estimation: Estimated absolute 3D pose from monocular images. Monocular 3D human pose estimation aims to recover the 3D joint coordinates of a person from a single 2D image, a task made inherently difficult by the loss of depth cues during image projection. To overcome the resulting depth ambiguity, we propose NovelPoseNet, a multi-modal ResNet-Transformer framework that explicitly generates novel 2D poses from unseen viewpoints conditioned on learned 2D pose features and hypothetical camera extrinsics. Unlike prior works that impose multi-view consistency as a supervision constraint, NovelPoseNet synthesizes explicit 2D pose representations from new camera perspectives, thereby enabling true multi-view triangulation from a single image. The reconstructed 3D pose is obtained by fusing these synthesized viewpoints through geometric triangulation, yielding both absolute and relative 3D joint positions. Extensive experiments on Human3.6M and MPI-INF-3DHP datasets demonstrate that NovelPoseNet achieves state-of-the-art performance, surpassing existing reprojection- and synthesis-based methods. [ABSTRACT FROM AUTHOR]
ISSN:01678655
DOI:10.1016/j.patrec.2026.02.007