Method

Object Proposals Re-ranking for Object Detection in Autonomous Driving [DeepStereoOP]


Submitted on 22 Feb. 2017 07:13 by
Cuong Pham (Sungkyunkwan University)

Running time:3.4 s
Environment:GPU @ 3.5 Ghz (Matlab + C/C++)

Method Description:
Re-ranking class-independent 3DOP object proposals
using a two-stream CNN with RGB and depth features.
Parameters:
-
Latex Bibtex:
@article{Pham2017SPIC,
author = {Cuong Cao Pham and Jae Wook Jeon},
title = {Robust Object Proposals Re-ranking for
Object Detection in Autonomous Driving Using
Convolutional Neural Networks},
journal = {Signal Processing: Image
Communiation},
year = {2017}
}

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) 90.34 % 88.75 % 79.39 %
Car (Orientation) 89.01 % 86.57 % 77.13 %
Pedestrian (Detection) 82.50 % 67.32 % 65.14 %
Pedestrian (Orientation) 73.37 % 59.28 % 56.87 %
Cyclist (Detection) 77.00 % 65.72 % 57.74 %
Cyclist (Orientation) 67.49 % 55.62 % 48.85 %
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