JTL-Net: joint task learning network for image-based localization

Kaixuan Men; Ruiming Jia; Xin Chen; Jiali Cui

doi:10.1117/12.2639712

19 October 2022 JTL-Net: joint task learning network for image-based localization

Kaixuan Men, Ruiming Jia, Xin Chen, Jiali Cui

Proceedings Volume 12294, 7th International Symposium on Advances in Electrical, Electronics, and Computer Engineering; 122944H (2022) https://doi.org/10.1117/12.2639712
Event: 7th International Symposium on Advances in Electrical, Electronics and Computer Engineering (ISAEECE 2022), 2022, Xishuangbanna, China

Abstract

To estimate the 6-DoF camera pose from a single RGB image, we propose a joint task learning (JTL) network called JTL-Net. JTL-Net is a convolutional neutral network with an asymmetric encoder–decoder structure consisting of a shared bone network as an encoder, two independent decoders, and a JTL regressor to output camera pose. Unlike most pose estimation networks, JTL-Net considers camera position and orientation as two specific streams to reduce the interference caused by differences between positions and angles. The JTL regressor, with an attention-guided architecture, is designed to deliver spatial information between these two streams. Furthermore, to improve pose accuracy, an auxiliary task stream is separated from the position stream to provide prior knowledge. We evaluated the effectiveness of our JTL-Net on benchmark datasets and found that it achieves great performance.

Citation Download Citation

Kaixuan Men, Ruiming Jia, Xin Chen, and Jiali Cui "JTL-Net: joint task learning network for image-based localization", Proc. SPIE 12294, 7th International Symposium on Advances in Electrical, Electronics, and Computer Engineering, 122944H (19 October 2022); https://doi.org/10.1117/12.2639712

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available