Method

VMVS [la] [VMVS]


Submitted on 28 Feb. 2019 19:43 by
Jason Ku (University of Toronto)

Running time:0.25 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
-
Parameters:
-
Latex Bibtex:
@inproceedings{ku2018joint,
title={Improving 3D object detection for
pedestrians with virtual multi-view synthesis
orientation estimation},
author={Ku, Jason and Pon, Alex D and Walsh,
Sean and Waslander, Steven L},
booktitle={IROS},
year={2019},
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
Pedestrian (Detection) 82.80 % 71.82 % 66.85 %
Pedestrian (Orientation) 79.98 % 68.19 % 63.18 %
Pedestrian (3D Detection) 53.44 % 43.27 % 39.51 %
Pedestrian (Bird's Eye View) 60.34 % 50.34 % 46.45 %
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