CaltechAUTHORS
  A Caltech Library Service

M²BEV: Multi-Camera Joint 3D Detection and Segmentation with Unified Birds-Eye View Representation

Xie, Enze and Yu, Zhiding and Zhou, Daquan and Philion, Jonah and Anandkumar, Anima and Fidler, Sanja and Luo, Ping and Alvarez, Jose M. (2022) M²BEV: Multi-Camera Joint 3D Detection and Segmentation with Unified Birds-Eye View Representation. . (Unpublished) https://resolver.caltech.edu/CaltechAUTHORS:20220714-212525848

[img] PDF - Submitted Version
See Usage Policy.

15MB

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20220714-212525848

Abstract

In this paper, we propose M$²BEV, a unified framework that jointly performs 3D object detection and map segmentation in the Birds Eye View~(BEV) space with multi-camera image inputs. Unlike the majority of previous works which separately process detection and segmentation, M$²BEV infers both tasks with a unified model and improves efficiency. M2BEV efficiently transforms multi-view 2D image features into the 3D BEV feature in ego-car coordinates. Such BEV representation is important as it enables different tasks to share a single encoder. Our framework further contains four important designs that benefit both accuracy and efficiency: (1) An efficient BEV encoder design that reduces the spatial dimension of a voxel feature map. (2) A dynamic box assignment strategy that uses learning-to-match to assign ground-truth 3D boxes with anchors. (3) A BEV centerness re-weighting that reinforces with larger weights for more distant predictions, and (4) Large-scale 2D detection pre-training and auxiliary supervision. We show that these designs significantly benefit the ill-posed camera-based 3D perception tasks where depth information is missing. M2BEV is memory efficient, allowing significantly higher resolution images as input, with faster inference speed. Experiments on nuScenes show that M$²BEV achieves state-of-the-art results in both 3D object detection and BEV segmentation, with the best single model achieving 42.5 mAP and 57.0 mIoU in these two tasks, respectively.


Item Type:Report or Paper (Discussion Paper)
Related URLs:
URLURL TypeDescription
https://doi.org/10.48550/arXiv.2207.05850arXivDiscussion Paper
https://xieenze.github.io/projects/m2bev/Related ItemProject website
ORCID:
AuthorORCID
Anandkumar, Anima0000-0002-6974-6797
Subject Keywords:Multi-Camera, Multi-Task Learning, Autonomous Driving
Record Number:CaltechAUTHORS:20220714-212525848
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20220714-212525848
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:115587
Collection:CaltechAUTHORS
Deposited By: George Porter
Deposited On:15 Jul 2022 22:39
Last Modified:15 Jul 2022 22:39

Repository Staff Only: item control page