Range Conditioned Dilated Convolutions for Scale Invariant 3D Object Detection

Alex Bewley, Pei Sun, Thomas Mensink, Dragomir Anguelov, Cristian Sminchisescu
Proceedings of the 2020 Conference on Robot Learning, PMLR 155:627-641, 2021.

Abstract

This paper presents a novel 3D object detection framework that processes LiDAR data directly on its native representation: range images. Benefiting from the compactness of range images, 2D convolutions can efficiently process dense LiDAR data of the scene. To overcome scale sensitivity in this perspective view, a novel range-conditioned dilation (RCD) layer is proposed to dynamically adjust a continuous dilation rate as a function of the measured range. Furthermore, localized soft range gating combined with a 3D box-refinement stage improves robustness in occluded areas, and produces overall more accurate bounding box predictions. On the public large-scale Waymo Open Dataset, our method sets a new baseline for range-based 3D detection, outperforming multiview and voxel-based methods over all ranges with unparalleled performance at long range detection.

Cite this Paper


BibTeX
@InProceedings{pmlr-v155-bewley21a, title = {Range Conditioned Dilated Convolutions for Scale Invariant 3D Object Detection}, author = {Bewley, Alex and Sun, Pei and Mensink, Thomas and Anguelov, Dragomir and Sminchisescu, Cristian}, booktitle = {Proceedings of the 2020 Conference on Robot Learning}, pages = {627--641}, year = {2021}, editor = {Kober, Jens and Ramos, Fabio and Tomlin, Claire}, volume = {155}, series = {Proceedings of Machine Learning Research}, month = {16--18 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v155/bewley21a/bewley21a.pdf}, url = {https://proceedings.mlr.press/v155/bewley21a.html}, abstract = {This paper presents a novel 3D object detection framework that processes LiDAR data directly on its native representation: range images. Benefiting from the compactness of range images, 2D convolutions can efficiently process dense LiDAR data of the scene. To overcome scale sensitivity in this perspective view, a novel range-conditioned dilation (RCD) layer is proposed to dynamically adjust a continuous dilation rate as a function of the measured range. Furthermore, localized soft range gating combined with a 3D box-refinement stage improves robustness in occluded areas, and produces overall more accurate bounding box predictions. On the public large-scale Waymo Open Dataset, our method sets a new baseline for range-based 3D detection, outperforming multiview and voxel-based methods over all ranges with unparalleled performance at long range detection.} }
Endnote
%0 Conference Paper %T Range Conditioned Dilated Convolutions for Scale Invariant 3D Object Detection %A Alex Bewley %A Pei Sun %A Thomas Mensink %A Dragomir Anguelov %A Cristian Sminchisescu %B Proceedings of the 2020 Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2021 %E Jens Kober %E Fabio Ramos %E Claire Tomlin %F pmlr-v155-bewley21a %I PMLR %P 627--641 %U https://proceedings.mlr.press/v155/bewley21a.html %V 155 %X This paper presents a novel 3D object detection framework that processes LiDAR data directly on its native representation: range images. Benefiting from the compactness of range images, 2D convolutions can efficiently process dense LiDAR data of the scene. To overcome scale sensitivity in this perspective view, a novel range-conditioned dilation (RCD) layer is proposed to dynamically adjust a continuous dilation rate as a function of the measured range. Furthermore, localized soft range gating combined with a 3D box-refinement stage improves robustness in occluded areas, and produces overall more accurate bounding box predictions. On the public large-scale Waymo Open Dataset, our method sets a new baseline for range-based 3D detection, outperforming multiview and voxel-based methods over all ranges with unparalleled performance at long range detection.
APA
Bewley, A., Sun, P., Mensink, T., Anguelov, D. & Sminchisescu, C.. (2021). Range Conditioned Dilated Convolutions for Scale Invariant 3D Object Detection. Proceedings of the 2020 Conference on Robot Learning, in Proceedings of Machine Learning Research 155:627-641 Available from https://proceedings.mlr.press/v155/bewley21a.html.

Related Material