BriefGPT.xyz
Dec, 2020
具有多个等变对齐的旋转不变点卷积
Rotation-Invariant Point Convolution With Multiple Equivariant Alignments
HTML
PDF
Hugues Thomas
TL;DR
研究了3D点卷积中平移不变性和等变性的关系，并使用旋转等变对齐改进了卷积网络，实现了物体分类和语义分割方面的表现提升，减少平移不变和标准3D深度学习方法之间的差距。
Abstract
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 net
→