LabelFormer: Object Trajectory Refinement for Offboard Perception from LiDAR Point Clouds

Anqi Joyce Yang, Sergio Casas, Nikita Dvornik, Sean Segal, Yuwen Xiong, Jordan Sir Kwang Hu, Carter Fang, Raquel Urtasun
Proceedings of The 7th Conference on Robot Learning, PMLR 229:3364-3383, 2023.

Abstract

A major bottleneck to scaling-up training of self-driving perception systems are the human annotations required for supervision. A promising alternative is to leverage “auto-labelling" offboard perception models that are trained to automatically generate annotations from raw LiDAR point clouds at a fraction of the cost. Auto-labels are most commonly generated via a two-stage approach – first objects are detected and tracked over time, and then each object trajectory is passed to a learned refinement model to improve accuracy. Since existing refinement models are overly complex and lack advanced temporal reasoning capabilities, in this work we propose LabelFormer, a simple, efficient, and effective trajectory-level refinement approach. Our approach first encodes each frame’s observations separately, then exploits self-attention to reason about the trajectory with full temporal context, and finally decodes the refined object size and per-frame poses. Evaluation on both urban and highway datasets demonstrates that LabelFormer outperforms existing works by a large margin. Finally, we show that training on a dataset augmented with auto-labels generated by our method leads to improved downstream detection performance compared to existing methods. Please visit the project website for details https://waabi.ai/labelformer/.

Cite this Paper

BibTeX
@InProceedings{pmlr-v229-yang23e, title = {LabelFormer: Object Trajectory Refinement for Offboard Perception from LiDAR Point Clouds}, author = {Yang, Anqi Joyce and Casas, Sergio and Dvornik, Nikita and Segal, Sean and Xiong, Yuwen and Hu, Jordan Sir Kwang and Fang, Carter and Urtasun, Raquel}, booktitle = {Proceedings of The 7th Conference on Robot Learning}, pages = {3364--3383}, year = {2023}, editor = {Tan, Jie and Toussaint, Marc and Darvish, Kourosh}, volume = {229}, series = {Proceedings of Machine Learning Research}, month = {06--09 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v229/yang23e/yang23e.pdf}, url = {https://proceedings.mlr.press/v229/yang23e.html}, abstract = {A major bottleneck to scaling-up training of self-driving perception systems are the human annotations required for supervision. A promising alternative is to leverage “auto-labelling" offboard perception models that are trained to automatically generate annotations from raw LiDAR point clouds at a fraction of the cost. Auto-labels are most commonly generated via a two-stage approach – first objects are detected and tracked over time, and then each object trajectory is passed to a learned refinement model to improve accuracy. Since existing refinement models are overly complex and lack advanced temporal reasoning capabilities, in this work we propose LabelFormer, a simple, efficient, and effective trajectory-level refinement approach. Our approach first encodes each frame’s observations separately, then exploits self-attention to reason about the trajectory with full temporal context, and finally decodes the refined object size and per-frame poses. Evaluation on both urban and highway datasets demonstrates that LabelFormer outperforms existing works by a large margin. Finally, we show that training on a dataset augmented with auto-labels generated by our method leads to improved downstream detection performance compared to existing methods. Please visit the project website for details https://waabi.ai/labelformer/.} }
Endnote
%0 Conference Paper %T LabelFormer: Object Trajectory Refinement for Offboard Perception from LiDAR Point Clouds %A Anqi Joyce Yang %A Sergio Casas %A Nikita Dvornik %A Sean Segal %A Yuwen Xiong %A Jordan Sir Kwang Hu %A Carter Fang %A Raquel Urtasun %B Proceedings of The 7th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Jie Tan %E Marc Toussaint %E Kourosh Darvish %F pmlr-v229-yang23e %I PMLR %P 3364--3383 %U https://proceedings.mlr.press/v229/yang23e.html %V 229 %X A major bottleneck to scaling-up training of self-driving perception systems are the human annotations required for supervision. A promising alternative is to leverage “auto-labelling" offboard perception models that are trained to automatically generate annotations from raw LiDAR point clouds at a fraction of the cost. Auto-labels are most commonly generated via a two-stage approach – first objects are detected and tracked over time, and then each object trajectory is passed to a learned refinement model to improve accuracy. Since existing refinement models are overly complex and lack advanced temporal reasoning capabilities, in this work we propose LabelFormer, a simple, efficient, and effective trajectory-level refinement approach. Our approach first encodes each frame’s observations separately, then exploits self-attention to reason about the trajectory with full temporal context, and finally decodes the refined object size and per-frame poses. Evaluation on both urban and highway datasets demonstrates that LabelFormer outperforms existing works by a large margin. Finally, we show that training on a dataset augmented with auto-labels generated by our method leads to improved downstream detection performance compared to existing methods. Please visit the project website for details https://waabi.ai/labelformer/.
APA
Yang, A.J., Casas, S., Dvornik, N., Segal, S., Xiong, Y., Hu, J.S.K., Fang, C. & Urtasun, R.. (2023). LabelFormer: Object Trajectory Refinement for Offboard Perception from LiDAR Point Clouds. Proceedings of The 7th Conference on Robot Learning, in Proceedings of Machine Learning Research 229:3364-3383 Available from https://proceedings.mlr.press/v229/yang23e.html.

Related Material