Method

Pseudo-LiDAR++ (SDN + GDC (4-beam LiDAR)) [st] [la] [PL++ (SDN+GDC)]
https://github.com/mileyan/Pseudo_Lidar_V2

Submitted on 31 Jul. 2019 17:08 by
Yurong You (Cornell University)

Running time:0.6 s
Environment:GPU @ 2.5 Ghz (C/C++)

Method Description:
Pseudo-LiDAR++ (SDN + GDC (4-beam LiDAR))
Parameters:
Pseudo-LiDAR++ (SDN + GDC (4-beam LiDAR))
Latex Bibtex:
@inproceedings{
you2020pseudolidar,
title={Pseudo-LiDAR++: Accurate Depth for 3D
Object Detection in Autonomous Driving},
author={Yurong You and Yan Wang and Wei-Lun
Chao and Divyansh Garg and Geoff Pleiss and
Bharath Hariharan and Mark Campbell and Kilian
Q.
Weinberger},
booktitle={International Conference on Learning
Representations},
year={2020},
url={https://openreview.net/forum?
id=BJedHRVtPB}
}

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) 94.95 % 85.15 % 77.78 %
Car (Orientation) 94.83 % 84.42 % 76.95 %
Car (3D Detection) 68.38 % 54.88 % 49.16 %
Car (Bird's Eye View) 84.61 % 73.80 % 65.59 %
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