Method

YOLOMono3D: Real-time Monocular 3D Object Detection [YoloMono3D]
https://github.com/Owen-Liuyuxuan/visualDet3D

Submitted on 30 Jun. 2020 08:17 by
Yuxuan LIU (Hong Kong University of Science and Technology)

Running time:0.05 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
This work follows the idea of M3D-RPN to directly
predict 3D anchors, and the code was completely
reconstructed to have a different definition of
the anchors. Improve on implementation of hill-
climbing algorithms to make inference much faster
(0.05s/pic for each image, including network
inference, post-optimization, and file IO) while
achieving a similar performance boost. U
Parameters:
Mono only.
Latex Bibtex:
@inproceedings{liu2021yolostereo3d,
title={YOLOStereo3D: A Step Back to 2D for
Efficient Stereo 3D Detection},
author={Yuxuan Liu and Lujia Wang and Ming, Liu},
booktitle={2021 International Conference on
Robotics and Automation (ICRA)},
year={2021},
organization={IEEE}
}

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) 92.37 % 79.63 % 59.69 %
Car (Orientation) 91.43 % 78.50 % 58.80 %
Car (3D Detection) 18.28 % 12.06 % 8.42 %
Car (Bird's Eye View) 26.79 % 17.15 % 12.56 %
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