Representation Learning for Object Detection from Unlabeled Point Cloud Sequences

Xiangru Huang, Yue Wang, Vitor Campagnolo Guizilini, Rares Andrei Ambrus, Adrien Gaidon, Justin Solomon
Proceedings of The 6th Conference on Robot Learning, PMLR 205:1277-1288, 2023.

Abstract

Although unlabeled 3D data is easy to collect, state-of-the-art machine learning techniques for 3D object detection still rely on difficult-to-obtain manual annotations. To reduce dependence on the expensive and error-prone process of manual labeling, we propose a technique for representation learning from unlabeled LiDAR point cloud sequences. Our key insight is that moving objects can be reliably detected from point cloud sequences without the need for human-labeled 3D bounding boxes. In a single LiDAR frame extracted from a sequence, the set of moving objects provides sufficient supervision for single-frame object detection. By designing appropriate pretext tasks, we learn point cloud features that generalize to both moving and static unseen objects. We apply these features to object detection, achieving strong performance on self-supervised representation learning and unsupervised object detection tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v205-huang23b, title = {Representation Learning for Object Detection from Unlabeled Point Cloud Sequences}, author = {Huang, Xiangru and Wang, Yue and Guizilini, Vitor Campagnolo and Ambrus, Rares Andrei and Gaidon, Adrien and Solomon, Justin}, booktitle = {Proceedings of The 6th Conference on Robot Learning}, pages = {1277--1288}, year = {2023}, editor = {Liu, Karen and Kulic, Dana and Ichnowski, Jeff}, volume = {205}, series = {Proceedings of Machine Learning Research}, month = {14--18 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v205/huang23b/huang23b.pdf}, url = {https://proceedings.mlr.press/v205/huang23b.html}, abstract = {Although unlabeled 3D data is easy to collect, state-of-the-art machine learning techniques for 3D object detection still rely on difficult-to-obtain manual annotations. To reduce dependence on the expensive and error-prone process of manual labeling, we propose a technique for representation learning from unlabeled LiDAR point cloud sequences. Our key insight is that moving objects can be reliably detected from point cloud sequences without the need for human-labeled 3D bounding boxes. In a single LiDAR frame extracted from a sequence, the set of moving objects provides sufficient supervision for single-frame object detection. By designing appropriate pretext tasks, we learn point cloud features that generalize to both moving and static unseen objects. We apply these features to object detection, achieving strong performance on self-supervised representation learning and unsupervised object detection tasks. } }
Endnote
%0 Conference Paper %T Representation Learning for Object Detection from Unlabeled Point Cloud Sequences %A Xiangru Huang %A Yue Wang %A Vitor Campagnolo Guizilini %A Rares Andrei Ambrus %A Adrien Gaidon %A Justin Solomon %B Proceedings of The 6th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Karen Liu %E Dana Kulic %E Jeff Ichnowski %F pmlr-v205-huang23b %I PMLR %P 1277--1288 %U https://proceedings.mlr.press/v205/huang23b.html %V 205 %X Although unlabeled 3D data is easy to collect, state-of-the-art machine learning techniques for 3D object detection still rely on difficult-to-obtain manual annotations. To reduce dependence on the expensive and error-prone process of manual labeling, we propose a technique for representation learning from unlabeled LiDAR point cloud sequences. Our key insight is that moving objects can be reliably detected from point cloud sequences without the need for human-labeled 3D bounding boxes. In a single LiDAR frame extracted from a sequence, the set of moving objects provides sufficient supervision for single-frame object detection. By designing appropriate pretext tasks, we learn point cloud features that generalize to both moving and static unseen objects. We apply these features to object detection, achieving strong performance on self-supervised representation learning and unsupervised object detection tasks.
APA
Huang, X., Wang, Y., Guizilini, V.C., Ambrus, R.A., Gaidon, A. & Solomon, J.. (2023). Representation Learning for Object Detection from Unlabeled Point Cloud Sequences. Proceedings of The 6th Conference on Robot Learning, in Proceedings of Machine Learning Research 205:1277-1288 Available from https://proceedings.mlr.press/v205/huang23b.html.

Related Material