Method

VGCN [VGCN]
[Anonymous Submission]

Submitted on 19 Jan. 2021 10:02 by
[Anonymous Submission]

Running time:0.09 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
A new 3D Point cloud object detection method based
on graph convolution network.
Voxelizing entire space in pre-process, then set
up graph structure of the point cloud for time
efficiency。
End-to-end object detection network。
Parameters:
alpha=1,beta=10,gama=0.2
Latex Bibtex:
Y. Zhou, O. Tuzel, Voxelnet: End-to-end
learning for point cloud based3d object detection,
in: Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, 2018, pp. 4490–
4499.
Yan, Yan, Yuxing Mao, and Bo Li. "Second:
Sparsely embedded convolutional detection."
Sensors 18.10 (2018): 3337.
KIPF, Thomas N.; WELLING, Max. Semi-supervised
classification with graph convolutional networks.
arXiv preprint arXiv:1609.02907, 2016.

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) 94.91 % 91.97 % 89.34 %
Car (Orientation) 94.88 % 91.80 % 89.06 %
Car (3D Detection) 84.47 % 77.65 % 73.36 %
Car (Bird's Eye View) 90.67 % 87.16 % 82.98 %
Pedestrian (Detection) 61.86 % 52.80 % 50.66 %
Pedestrian (Orientation) 57.79 % 48.42 % 45.87 %
Pedestrian (3D Detection) 45.28 % 37.60 % 34.96 %
Pedestrian (Bird's Eye View) 50.02 % 42.33 % 40.05 %
Cyclist (Detection) 86.81 % 72.28 % 65.68 %
Cyclist (Orientation) 86.42 % 71.56 % 65.01 %
Cyclist (3D Detection) 78.47 % 62.36 % 55.88 %
Cyclist (Bird's Eye View) 81.50 % 67.04 % 59.45 %
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



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



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