Method

Multi-view Deformable Part Model trained with structured bounding box and viewpoint loss [DPM-VOC+VP]


Submitted on 13 Mar. 2015 14:18 by
Bojan Pepikj (Max Planck Institute for Informatics)

Running time:8 s
Environment:1 core @ 2.5 Ghz (C/C++)

Method Description:
A multi-view Deformable Part Model, trained with a
structured output loss, addressing object
localization and viewpoint estimation.
Parameters:
C=.002
\alpha = 0.5
Latex Bibtex:
@Article{Pepik2015PAMI,
title = "Multi-view and 3D Deformable Part
Models",
journal = "IEEE Transactions on Pattern
Analysis and Machine Intelligence (TPAMI)",
year = "2015",
publisher = "IEEE",
type = "Journal",
author = "Bojan Pepik and Michael Stark
and Peter Gehler and Bernt Schiele"
}

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) 82.15 % 66.72 % 49.01 %
Car (Orientation) 79.09 % 63.58 % 46.59 %
Pedestrian (Detection) 59.21 % 43.26 % 38.12 %
Pedestrian (Orientation) 52.91 % 37.79 % 33.27 %
Cyclist (Detection) 41.58 % 27.73 % 24.61 %
Cyclist (Orientation) 27.97 % 18.92 % 17.43 %
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



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



Orientation estimation results.
This figure as: png eps pdf txt gnuplot



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