Method

Self-supervised PointPillars [SSL_PP]
[Anonymous Submission]

Submitted on 1 Jun. 2022 23:40 by
[Anonymous Submission]

Running time:16ms
Environment:GPU @ 1.5 Ghz (Python)

Method Description:
We propose a self-supervised (SSL) generic point cloud pre-training
Parameters:
same as mmdetection3d default setting
Latex Bibtex:

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) 95.97 % 92.04 % 84.91 %
Car (Orientation) 95.62 % 91.31 % 84.09 %
Car (3D Detection) 83.74 % 72.02 % 64.95 %
Car (Bird's Eye View) 92.19 % 85.93 % 80.40 %
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