Method

Patch - Ensemble Multiple Proposals [la] [Patches - EMP]


Submitted on 28 Feb. 2019 05:21 by
Johannes Lehner (JKU Linz)

Running time:0.5 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
An ensemble version of our model. The predictions
from three models are aggregated. Then
Non-Maximum Suppression with an IoU-threshold of
0.7 is applied.
Parameters:
TBD
Latex Bibtex:
@article{lehner2019patch,
title={Patch Refinement: Localized 3D
Object Detection},
author={Johannes Lehner and Andreas
Mitterecker and Thomas Adler and Markus
Hofmarcher and Bernhard Nessler and Sepp
Hochreiter},
journal={arXiv preprint arXiv:1910.04093},
year={2019},
}

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) 97.91 % 93.75 % 90.56 %
Car (Orientation) 97.88 % 93.58 % 90.31 %
Car (3D Detection) 89.84 % 78.41 % 73.15 %
Car (Bird's Eye View) 94.49 % 88.17 % 84.75 %
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