Method

Single-Stage Monocular 3D Object Detection with LiDAR-Compatible Depth Prediction [LCD3D]
[Anonymous Submission]

Submitted on 5 Jun. 2020 21:46 by
[Anonymous Submission]

Running time:0.03 s
Environment:GPU @ 2.5 Ghz (Python)

Method Description:
LCD3D is based on a monocular camera based method.
The inference is performed via predicting depth on the surface of the object, which is similar to the depth information obtained by LiDAR.
Therefore, in situations where LiDAR data are available, LCD3D can replace the predicted depth with the LiDAR-estimated depth for more accurate object detection.
Parameters:
th_pc: 0.4
without LiDAR
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) 91.29 % 81.25 % 64.55 %
Car (Orientation) 91.20 % 81.01 % 64.29 %
Car (3D Detection) 13.77 % 9.04 % 7.23 %
Car (Bird's Eye View) 21.97 % 13.99 % 11.43 %
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