Rotation-Invariant Point Convolution With Multiple Equivariant Alignments
This work addresses the performance gap between rotation-invariant and standard 3D deep learning approaches for researchers and practitioners working with 3D data.
This paper explores the relationship between equivariance and invariance in 3D point convolutions, demonstrating that rotation-equivariant alignments can make any convolutional layer rotation-invariant. The authors enhance this by integrating alignments as features and combining multiple alignments, leading to state-of-the-art results in object classification and semantic segmentation.
Recent attempts at introducing rotation invariance or equivariance in 3D deep learning approaches have shown promising results, but these methods still struggle to reach the performances of standard 3D neural networks. In this work we study the relation between equivariance and invariance in 3D point convolutions. We show that using rotation-equivariant alignments, it is possible to make any convolutional layer rotation-invariant. Furthermore, we improve this simple alignment procedure by using the alignment themselves as features in the convolution, and by combining multiple alignments together. With this core layer, we design rotation-invariant architectures which improve state-of-the-art results in both object classification and semantic segmentation and reduces the gap between rotation-invariant and standard 3D deep learning approaches.