NeuRAD: Neural Rendering for Autonomous Driving
Paper in proceeding, 2024
Neural radiance fields (NeRFs) have gained popularity in the autonomous driving (AD) community. Recent meth-ods show NeRFs' potential for closed-loop simulation, en-abling testing of AD systems, and as an advanced training data augmentation technique. However, existing meth-ods often require long training times, dense semantic su-pervision, or lack generalizability. This, in turn, hinders the application of NeRFs for AD at scale. In this paper, we propose NeuRAD, a robust novel view synthesis method tailored to dynamic AD data. Our method features simple network design, extensive sensor modeling for both cam-era and lidar - including rolling shutter, beam divergence and ray dropping - and is applicable to multiple datasets out of the box. We verify its performance on five popular AD datasets, achieving state-of-the-art performance across the board. To encourage further development, we openly release the NeuRAD source code.
Autonomous Driving
NeRF
Neural Rendering
Author
Adam Tonderski
Zenseact AB
Lund University
Carl Lindström
Chalmers, Electrical Engineering, Signal Processing and Biomedical Engineering
Zenseact AB
Georg Hess
Zenseact AB
Chalmers, Electrical Engineering, Signal Processing and Biomedical Engineering
William Ljungbergh
Zenseact AB
Linköping University
Lennart Svensson
Chalmers, Electrical Engineering, Signal Processing and Biomedical Engineering
Christoffer Petersson
Chalmers, Mathematical Sciences, Algebra and geometry
Zenseact AB
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
10636919 (ISSN)
14895-14904
2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024
Seattle, USA,
Seattle, USA,
Subject Categories (SSIF 2025)
Computer graphics and computer vision
Computer Sciences
DOI
10.1109/CVPR52733.2024.01411