Method

Voxel R-CNN [Voxel R-CNN]
https://github.com/djiajunustc/Voxel-R-CNN

Submitted on 8 Sep. 2020 09:15 by
Jiajun Deng (University of Science and Technology of China)

Running time:0.04 s
Environment:GPU @ 3.0 Ghz (C/C++)

Method Description:
TBD
Parameters:
TBD
Latex Bibtex:
@inproceedings{deng2020voxelrcnn,
title={Voxel R-CNN: Towards High Performance
Voxel-based 3D Object Detection
},
author={Deng, Jiajun and Shi, Shaoshuai and Li,
Peiwei and Zhou, Wengang and Zhang, Yanyong and
Li, Houqiang},
booktitle={AAAI},
year={2021}
}

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) 96.49 % 95.11 % 92.45 %
Car (Orientation) 96.47 % 94.96 % 92.24 %
Car (3D Detection) 90.90 % 81.62 % 77.06 %
Car (Bird's Eye View) 94.85 % 88.83 % 86.13 %
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