Method

Monocular 3D Object Detection With Fake 3Dbox For Autonomous Driving [MF3D]
[Anonymous Submission]

Submitted on 16 Oct. 2018 08:39 by
[Anonymous Submission]

Running time:0.03 s
Environment:GPU @ 2.5 Ghz (C/C++)

Method Description:
We present a novel approach for 3D object detection from a single image for autopilot. While the depth information is missing, we divide the task into two independent sub-problems according to point-line-face-cuboid principle.
Parameters:
monocular
Latex Bibtex:
N/A

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) 88.46 % 68.72 % 58.70 %
Car (Orientation) 87.79 % 67.68 % 57.57 %
Car (3D Detection) 3.81 % 3.17 % 3.25 %
Car (Bird's Eye View) 7.88 % 5.57 % 5.08 %
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




eXTReMe Tracker