Method

Mobile Stereo R-CNN [st] [Mobile Stereo R-CNN]


Submitted on 30 Mar. 2021 11:04 by
Mohamed Hussein (Faculty of Engineering, Ain Shams University)

Running time:1.8 s
Environment:NVIDIA Jetson TX2

Method Description:
This model is based on Stereo R-CNN and adjusted to
be more suitable for running on embedded target
NVIDIA Jetson TX2. It uses only stereo images in
both training and inference.
Parameters:
To be filled later
Latex Bibtex:
@inproceedings{mobilestereorcnn2021,
title={3D Object Detection using Mobile Stereo R-
CNN on Nvidia Jetson TX2},
author={Hussein, Mohamed K. and Khalil, Mahmoud
I. and Abdullah, Bassem A.},
booktitle={International Conference on Advanced
Engineering, Technology and Applications
(ICAETA)},
year={2021},
url=
{https://icaeta.aiplustech.org/assets/docs/Procee
dings_Book.pdf}
}

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) 90.08 % 76.73 % 62.23 %
Car (Orientation) 88.80 % 74.13 % 59.84 %
Car (3D Detection) 26.97 % 17.04 % 13.26 %
Car (Bird's Eye View) 44.51 % 28.78 % 22.30 %
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