Method

Guiding Network [GN]


Submitted on 22 Aug. 2016 21:39 by
Sang-Il Jung (IIP, POSTECH)

Running time:1 s
Environment:GPU @ 2.5 Ghz (Matlab + C/C++)

Method Description:
We propose a guiding network to assist with training a deep convolutional neural network (DCNN) to improve the accuracy of pedestrian detection. The guiding network is adaptively appended to the pedestrian region of the last convolutional layer; the guiding network helps the DCNN to learn the convolutional layers for pedestrian features by focusing on the pedestrian region. The guiding network is used only for training, and therefore does not affect the inference speed.
Parameters:
.
Latex Bibtex:
@article{JUNG201743,
title = "Deep network aided by guiding network for pedestrian detection",
journal = "Pattern Recognition Letters",
volume = "90",
number = "Supplement C",
pages = "43 - 49",
year = "2017",
issn = "0167-8655",
doi = "https://doi.org/10.1016/j.patrec.2017.02.018",
url = "http://www.sciencedirect.com/science/article/pii/S0167865517300545",
author = "Sang-Il Jung and Ki-Sang Hong",
keywords = "Pedestrian detection, Deep convolutional neural network"
}

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) 82.93 % 72.29 % 65.56 %
This table as LaTeX


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




eXTReMe Tracker