Adapting multi-task perception in autonomous driving to bird’s-eye-view representation via multi-level guidance

Yu Tian; ChangMing Zhu

doi:10.1117/12.2685837

15 August 2023 Adapting multi-task perception in autonomous driving to bird’s-eye-view representation via multi-level guidance

Yu Tian, ChangMing Zhu

Author Affiliations +

Proceedings Volume 12719, Second International Conference on Electronic Information Technology (EIT 2023); 127194Q (2023) https://doi.org/10.1117/12.2685837
Event: Second International Conference on Electronic Information Technology (EIT 2023), 2023, Wuhan, China

Abstract

Perception models using multi-camera images show promise in autonomous driving. This paper proposes a multi-task network leveraging these images and bird's eye view (BEV) perspective for efficient perception. While Transformers in BEV tasks help focus attention, they often require high computational cost. To address this, we introduce multi-level guidance in the Encoder and Decoder, using richer semantic information to initialize the Transformer for faster convergence and improved performance.We guide the initialization of 3D BEV features using 2D image features in the Encoder, and apply perspective supervision in the Decoder for optimized detection and segmentation. The detection head output guides the segmentation head to focus on regions of interest (ROI). Our method, validated on the nuScenes dataset, yields excellent results (54.8% NDS, 45.2% mAP, 49.7% mIoU). Code release is forthcoming.

Citation Download Citation

Yu Tian and ChangMing Zhu "Adapting multi-task perception in autonomous driving to bird’s-eye-view representation via multi-level guidance", Proc. SPIE 12719, Second International Conference on Electronic Information Technology (EIT 2023), 127194Q (15 August 2023); https://doi.org/10.1117/12.2685837

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

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
7 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Object detection

Image segmentation

Neodymium

3D image processing

Autonomous driving

Transformers

Feature extraction

Show All Keywords

Keywords/Phrases

Search In:

Publication Years