Method

ImVoxelNet [ImVoxelNet]
https://github.com/saic-vul/imvoxelnet

Submitted on 5 May. 2021 09:59 by
Anna Vorontsova (Samsung AI Center Russia)

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

Method Description:
Novel image to voxels projection method for indoor/outdoor monocular/multiview 3d object detection.
Parameters:
-
Latex Bibtex:
@article{rukhovich2021imvoxelnet,
title={ImVoxelNet: Image to Voxels Projection for Monocular and Multi-View General-Purpose 3D Object Detection},
author={Rukhovich, Danila and Vorontsova, Anna and Konushin, Anton},
journal={arXiv preprint arXiv:2106.01178},
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) 89.80 % 79.09 % 69.45 %
Car (Orientation) 89.07 % 77.18 % 67.35 %
Car (3D Detection) 17.15 % 10.97 % 9.15 %
Car (Bird's Eye View) 25.19 % 16.37 % 13.58 %
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