Method

PI-RCNN [PI-RCNN]


Submitted on 18 Nov. 2019 12:33 by
Liang Xie (ZJU)

Running time:0.1 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
TBD
Parameters:
TBD
Latex Bibtex:
@inproceedings{xie2020pi,
title="PI-RCNN: An Efficient Multi-sensor 3D
Object Detector with Point-based Attentive Cont-conv
Fusion Module",
author="Liang {Xie} and Chao {Xiang} and
Zhengxu {Yu} and Guodong {Xu} and Zheng {Yang} and
Deng {Cai} and Xiaofei {He}",
booktitle="AAAI 2020 : The Thirty-Fourth
AAAI Conference on Artificial Intelligence",
notes="Sourced from Microsoft Academic -
https://academic.microsoft.com/paper/2998254148",
year="2020"
}

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) 96.17 % 92.66 % 87.68 %
Car (Orientation) 96.15 % 92.52 % 87.47 %
Car (3D Detection) 84.37 % 74.82 % 70.03 %
Car (Bird's Eye View) 91.44 % 85.81 % 81.00 %
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