Method

Fitness Quality Network [FQNet]


Submitted on 8 Feb. 2019 07:02 by
Lijie Liu (Tsinghua University)

Running time:0.5 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
Fitness Quality Network
Parameters:
using train/val of Mono3D
Latex Bibtex:
@inproceedings{liu2019deep,
title={Deep Fitting Degree Scoring Network for
Monocular 3D Object Detection},
author={Liu, Lijie and Lu, Jiwen and Xu, Chunjing
and Tian, Qi and Zhou, Jie},
booktitle={Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition},
pages={1057--1066},
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) 94.72 % 90.17 % 76.78 %
Car (Orientation) 93.66 % 87.49 % 73.61 %
Car (3D Detection) 2.77 % 1.51 % 1.01 %
Car (Bird's Eye View) 5.40 % 3.23 % 2.46 %
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