KITTI-360

Method

MinkowskiNet Res16UNet34 [MinkowskiNet]
https://github.com/drprojects/DeepViewAgg

Submitted on 7 Feb. 2022 18:24 by
Damien Robert (Institut Géographique National)

Running time:-
Environment:NVIDIA V100

Method Description:
Implementation of MinkowskiNet from the Torch-Points3D
library. Operates on cylindrical samplings of the colorized
point cloud.
Parameters:
Res16UNet34 architecture, cyilnder_radius=6
Latex Bibtex:
@inproceedings{choy20194d,
title={4d spatio-temporal convnets: Minkowski
convolutional neural
networks},
author={Choy, Christopher and Gwak, JunYoung and
Savarese, Silvio},
booktitle={Proceedings of the IEEE/CVF Conference on
Computer
Vision and Pattern Recognition},
pages={3075--3084},
year={2019}
}

@inproceedings{robert2022dva,
title={Learning Multi-View Aggregation In the Wild for
Large-Scale 3D Semantic Segmentation},
author={Robert, Damien and Vallet, Bruno and
Landrieu, Loic},
booktitle={Proceedings of the IEEE/CVF Conference on
Computer Vision and Pattern Recognition},
year={2022}
}

Test Set Average

road sidewalk building wall fence pole traffic light traffic sign vegetation terrain person car truck motorcycle bicycle mIoU class
94.17 81.37 85.99 45.21 44.42 55.07 0.00 37.23 90.20 85.26 20.28 96.34 53.12 6.55 13.54 53.92
flat construction object nature human vehicle mIoU category
95.82 81.27 58.87 92.20 20.28 96.07 74.08
This table as LaTeX




eXTReMe Tracker