在表面上使用旋转等变特征的符号纸cnn的规范

CNNs on Surfaces using Rotation-Equivariant Features Code for Harmonic Surface Networks, an approach for deep learning on surfaces operating on vector-valued, rotation-equivariant features. This is achieved by learning circular harmonic kernels and separating features in streams of different equivariance classes. The advantage of our approach is that the rotational degree of freedom, arising when