3D-Rotation-Equivariant Quaternion Neural Networks
Computer Vision and Pattern Recognition
2020-10-13 v2 Machine Learning
Machine Learning
Abstract
This paper proposes a set of rules to revise various neural networks for 3D point cloud processing to rotation-equivariant quaternion neural networks (REQNNs). We find that when a neural network uses quaternion features under certain conditions, the network feature naturally has the rotation-equivariance property. Rotation equivariance means that applying a specific rotation transformation to the input point cloud is equivalent to applying the same rotation transformation to all intermediate-layer quaternion features. Besides, the REQNN also ensures that the intermediate-layer features are invariant to the permutation of input points. Compared with the original neural network, the REQNN exhibits higher rotation robustness.
Keywords
Cite
@article{arxiv.1911.09040,
title = {3D-Rotation-Equivariant Quaternion Neural Networks},
author = {Wen Shen and Binbin Zhang and Shikun Huang and Zhihua Wei and Quanshi Zhang},
journal= {arXiv preprint arXiv:1911.09040},
year = {2020}
}