Invariance to spatial transformations such as translations and rotations is a
desirable property and a basic design principle for classification neural
networks. However, the commonly used convolutional neural networks (CNNs) are
actually very sensitive to even small translations. Ther
介绍了一种 3D 旋转等变 CNN (CubeNet),该网络通过保留 3D 形状的全局和局部特征,有助于维护体素化对象的有意义表示,并能解释输入之间的姿态差异。应用于各种 3D 推断问题中,在 ModelNet10 分类挑战赛上实现了最先进的性能,并在 ISBI 2012 Connectome 分割基准测试中实现了可比性能。