Method

ResNet18-based Recurrent Rolling Convolution (Car only) [ResNet-RRC_Car]


Submitted on 3 Jul. 2018 10:23 by
Hyung-Joon Jeon (Sungkyunkwan University)

Running time:0.06 s
Environment:GPU @ 1.5 Ghz (Python + C/C++)

Method Description:
This is the result entry for ResNet18-based
Recurrent Rolling Convolution.
Note that this model only aims to detect cars in
traffic, but not pedestrians and cyclists, thereby
having faster detection rates.
Parameters:
\alpha=0.0005
\mu=0.9
step_size=25000
learning_rate_decay=0.5
Number_of_Iterations=100000
Latex Bibtex:
@inproceedings{rrc-resnet,
author = {Jeon, Hyung-Joon and others},
title = {High-Speed Car Detection Using ResNet-
Based Recurrent Rolling Convolution},
booktitle = {Proceedings of the IEEE conference
on
systems, man, and cybernetics},
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) 91.45 % 85.33 % 74.27 %
This table as LaTeX

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




eXTReMe Tracker