Method

Learning Object Sub-Categories [SubCat]
http://cvrr.ucsd.edu/eshed

Submitted on 13 Mar. 2015 04:49 by
Eshed Ohn-Bar (UCSD)

Running time:0.7 s
Environment:6 cores @ 3.5 Ghz (Matlab + C/C++)

Method Description:
This is an updated version of the SubCat detector
Parameters:
See the associated code/publication. The main
difference is in learning a 3-resolutions model
and
downsampling feature channels by a factor of 2
instead of 4. Speed is shown for soft cascade
threshold of -1.
Latex Bibtex:
@article{Ohn-Bar2015TITS,
author = {Eshed Ohn-Bar and Mohan M. Trivedi},
title = {Learning to Detect Vehicles by
Clustering
Appearance Patterns},
journal = {T-ITS},
year = {2015},
url =
{cvrr.ucsd.edu/eshed/papers/OhnBar_TITS15_wsupp
lement.pdf},
}

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) 84.10 % 76.36 % 60.56 %
Car (Orientation) 83.31 % 75.26 % 59.55 %
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




eXTReMe Tracker