Stereo CenterNet[st] [Stereo CenterNet]

Submitted on 4 Dec. 2021 13:17 by
YuGuang Shi (University of Science and Technology Beijing)

Running time:0.04 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
Recently, three-dimensional (3D) detection based on
stereo images has progressed remarkably; however,
most advanced methods adopt anchor-based two-
dimensional (2D) detection or depth estimation to
address this problem. Nevertheless, high
computational cost inhibits these methods from
achieving real-time performance. In this study, we
propose a 3D object detection method, Stereo
CenterNet (SC), using geometric information in
stereo imagery. SC predicts the four semantic key
points of the 3D bounding box of the object in space
and utilizes 2D left and right boxes, 3D dimension,
orientation, and key points to restore the bounding
box of the object in the 3D space. Subsequently, we
adopt an improved photometric alignment module to
further optimize the position of the 3D bounding
box. Experiments conducted on the KITTI dataset
indicate that the proposed SC exhibits the best
speed-accuracy trade-off among advanced methods
without using extra data.
Latex Bibtex:
title = {Stereo CenterNet-based 3D object
detection for autonomous driving},
journal = {Neurocomputing},
volume = {471},
pages = {219-229},
year = {2022},
issn = {0925-2312},
doi =
url =
author = {Yuguang Shi and Yu Guo and
Zhenqiang Mi and Xinjie Li},


Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).

Benchmark Easy Moderate Hard
Car (Detection) 96.61 % 91.27 % 83.50 %
Car (Orientation) 96.54 % 91.02 % 83.15 %
Car (3D Detection) 49.94 % 31.30 % 25.62 %
Car (Bird's Eye View) 62.97 % 42.12 % 35.37 %
This table as LaTeX

2D object detection results.
This figure as: png eps pdf txt gnuplot

Orientation estimation results.
This figure as: png eps pdf txt gnuplot

3D object detection results.
This figure as: png eps pdf txt gnuplot

Bird's eye view results.
This figure as: png eps pdf txt gnuplot

eXTReMe Tracker