Method

Pillar-Based Pseudo-LiDAR Transfer for Monocular 3D Object Detection [PPTrans]
[Anonymous Submission]

Submitted on 28 May. 2021 14:42 by
[Anonymous Submission]

Running time:0.2 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
We propose a pillar-based framework that jointly
utilizes image and pseudo-LiDAR for scene-level 3D
object detection.
Parameters:
weight = 0.3
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) 95.17 % 89.17 % 81.77 %
Car (Orientation) 95.04 % 88.70 % 81.15 %
Car (3D Detection) 19.79 % 12.06 % 10.48 %
Car (Bird's Eye View) 28.05 % 18.12 % 15.41 %
Pedestrian (Detection) 45.02 % 34.45 % 30.94 %
Pedestrian (Orientation) 30.13 % 22.84 % 20.39 %
Pedestrian (3D Detection) 2.68 % 1.85 % 1.44 %
Pedestrian (Bird's Eye View) 4.01 % 2.58 % 2.37 %
Cyclist (Detection) 48.93 % 39.55 % 33.74 %
Cyclist (Orientation) 39.90 % 30.31 % 26.28 %
Cyclist (3D Detection) 2.31 % 1.38 % 1.20 %
Cyclist (Bird's Eye View) 3.44 % 2.07 % 1.77 %
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