Method

Filtered Channel Features using Checkerboards filter bank [FilteredICF]


Submitted on 6 Jul. 2015 12:57 by
Shanshan Zhang (MPI Informatik)

Running time:~ 2 s
Environment:>8 cores @ 2.5 Ghz (Matlab + C/C++)

Method Description:
It uses filtered channel features by applying Checkerboards4x3 filters on HOG+LUV. The classifier is a boosted decision forest, trained on top of the filtered channel features.
The running time is based on our un-optimized Matlab/C++ implementation.
Link to the paper: http://arxiv.org/abs/1501.05759
Parameters:
See the paper for parameters.
Latex Bibtex:
@INPROCEEDINGS{Zhang2015CVPR,
author = {Shanshan Zhang and Rodrigo Benenson and Bernt Schiele},
title = {Filtered Channel Features for Pedestrian Detection},
booktitle = CVPR,
year = {2015}
}

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) 69.79 % 56.53 % 50.32 %
This table as LaTeX

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




eXTReMe Tracker