Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations

Minghan Zhu, Maani Ghaffari, Huei Peng
Proceedings of the 5th Conference on Robot Learning, PMLR 164:1412-1422, 2022.

Abstract

This paper proposes a correspondence-free method for point cloud rotational registration. We learn an embedding for each point cloud in a feature space that preserves the SO(3)-equivariance property, enabled by recent developments in equivariant neural networks. The proposed shape registration method achieves three major advantages through combining equivariant feature learning with implicit shape models. First, the necessity of data association is removed because of the permutation-invariant property in network architectures similar to PointNet. Second, the registration in feature space can be solved in closed-form using Horn’s method due to the SO(3)-equivariance property. Third, the registration is robust to noise in the point cloud because of the joint training of registration and implicit shape reconstruction. The experimental results show superior performance compared with existing correspondence-free deep registration methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v164-zhu22b, title = {Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations}, author = {Zhu, Minghan and Ghaffari, Maani and Peng, Huei}, booktitle = {Proceedings of the 5th Conference on Robot Learning}, pages = {1412--1422}, year = {2022}, editor = {Faust, Aleksandra and Hsu, David and Neumann, Gerhard}, volume = {164}, series = {Proceedings of Machine Learning Research}, month = {08--11 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v164/zhu22b/zhu22b.pdf}, url = {https://proceedings.mlr.press/v164/zhu22b.html}, abstract = {This paper proposes a correspondence-free method for point cloud rotational registration. We learn an embedding for each point cloud in a feature space that preserves the SO(3)-equivariance property, enabled by recent developments in equivariant neural networks. The proposed shape registration method achieves three major advantages through combining equivariant feature learning with implicit shape models. First, the necessity of data association is removed because of the permutation-invariant property in network architectures similar to PointNet. Second, the registration in feature space can be solved in closed-form using Horn’s method due to the SO(3)-equivariance property. Third, the registration is robust to noise in the point cloud because of the joint training of registration and implicit shape reconstruction. The experimental results show superior performance compared with existing correspondence-free deep registration methods. } }
Endnote
%0 Conference Paper %T Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations %A Minghan Zhu %A Maani Ghaffari %A Huei Peng %B Proceedings of the 5th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2022 %E Aleksandra Faust %E David Hsu %E Gerhard Neumann %F pmlr-v164-zhu22b %I PMLR %P 1412--1422 %U https://proceedings.mlr.press/v164/zhu22b.html %V 164 %X This paper proposes a correspondence-free method for point cloud rotational registration. We learn an embedding for each point cloud in a feature space that preserves the SO(3)-equivariance property, enabled by recent developments in equivariant neural networks. The proposed shape registration method achieves three major advantages through combining equivariant feature learning with implicit shape models. First, the necessity of data association is removed because of the permutation-invariant property in network architectures similar to PointNet. Second, the registration in feature space can be solved in closed-form using Horn’s method due to the SO(3)-equivariance property. Third, the registration is robust to noise in the point cloud because of the joint training of registration and implicit shape reconstruction. The experimental results show superior performance compared with existing correspondence-free deep registration methods.
APA
Zhu, M., Ghaffari, M. & Peng, H.. (2022). Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations. Proceedings of the 5th Conference on Robot Learning, in Proceedings of Machine Learning Research 164:1412-1422 Available from https://proceedings.mlr.press/v164/zhu22b.html.

Related Material