Method

M3D-RPN: Monocular 3D Region Proposal Network for Object Detection [M3D-RPN]
http://cvlab.cse.msu.edu/project-m3d-rpn.html

Submitted on 11 Aug. 2019 14:56 by
Garrick Brazil (Michigan State University)

Running time:0.16 s
Environment:GPU @ 1.5 Ghz (Python)

Method Description:
We present a monocular 3D region proposal network
which performs multi-class 3D object detection
using a shared and single-shot network. For details
please refer to the paper.
Parameters:
See project website/paper/code for details.
Latex Bibtex:

@inproceedings{ brazil2019m3drpn,
author = { Brazil, Garrick and Liu, Xiaoming },
title = { M3D-RPN: Monocular 3D Region Proposal
Network for Object Detection },
booktitle = { ICCV },
address = { Seoul, South Korea },
month = { October },
year = { 2019 },
organization = { IEEE }
}

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) 89.04 % 85.08 % 69.26 %
Car (Orientation) 88.38 % 82.81 % 67.08 %
Car (3D Detection) 14.76 % 9.71 % 7.42 %
Car (Bird's Eye View) 21.02 % 13.67 % 10.23 %
Pedestrian (Detection) 56.64 % 41.46 % 37.31 %
Pedestrian (Orientation) 44.33 % 31.88 % 28.55 %
Pedestrian (3D Detection) 4.92 % 3.48 % 2.94 %
Pedestrian (Bird's Eye View) 5.65 % 4.05 % 3.29 %
Cyclist (Detection) 61.54 % 41.54 % 35.23 %
Cyclist (Orientation) 48.11 % 31.09 % 26.10 %
Cyclist (3D Detection) 0.94 % 0.65 % 0.47 %
Cyclist (Bird's Eye View) 1.25 % 0.81 % 0.78 %
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



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view 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



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view 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



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker