Method

RoarNet[la] [RoarNet]
https://github.com/Kiwoo/KazuaNet

Submitted on 21 Jul. 2018 01:25 by
Kiwoo Shin (UC Berkeley)

Running time:0.1 s
Environment:GPU @ >3.5 Ghz (Python + C/C++)

Method Description:
Collaborated research by Kiwoo Shin from
[Mechanical Systems Control lab@UC Berkeley &
Berkeley Deep Drive], Youngwook Paul Kwon from
[PhantomAI, UC Berkeley]

https://arxiv.org/pdf/1811.03818.pdf

Kiwoo Shin: kiwoo.shin@berkeley.edu
Youngwook Paul Kwon: young@berkeley.edu
Parameters:
N/A
Latex Bibtex:
@article{kiwoo2018roar,
title={RoarNet: A Robust 3D Object Detection based
on RegiOn Approximation Refinement},
author={Kiwoo Shin and YP Kwon and
Masayoshi Tomizuka},
journal={arXiv preprint arXiv:1811.03818},
year={2018}
}

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) 90.69 % 88.80 % 79.46 %
Car (3D Detection) 83.71 % 73.04 % 59.16 %
Car (Bird's Eye View) 88.20 % 79.41 % 70.02 %
This table as LaTeX


2D object detection 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