Method

Monocular 3D Object Detection with Feature Enhancement Networks [st] [StereoFENet]


Submitted on 7 May. 2019 04:39 by
Wentao Bao (Wuhan University)

Running time:0.15 s
Environment:1 core @ 3.5 Ghz (Python)

Method Description:
An extension of our previous CVPR2018 paper:
Multi-Level Fusion based 3D Object Detection from
Monocular Images.
Parameters:
N/A
Latex Bibtex:
@article{monofenet,
title={MonoFENet: Monocular 3D Object Detection
with
Feature Enhancement Networks},
author={Wentao Bao and Bin Xu and Zhenzhong
Chen},
journal={IEEE Transactions on Image Processing},
doi={10.1109/TIP.2019.2952201},
year={2019}
}

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) 91.48 % 85.70 % 77.62 %
Car (Orientation) 91.28 % 85.14 % 76.80 %
Car (3D Detection) 29.14 % 18.41 % 14.20 %
Car (Bird's Eye View) 49.29 % 32.96 % 25.90 %
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