Subdivision-based Mesh Convolution Networks

Convolutional neural networks (CNNs) have made great breakthroughs in 2D computer vision. However, their irregular structure makes it hard to harness the potential of CNNs directly on meshes. A subdivision surface provides a hierarchical multi-resolution structure, which each face closed 2-manifold triangle mesh is exactly adjacent three faces. Motivated by these two observations, this paper presents SubdivNet, an innovative and versatile CNN framework for 3D meshes with Loop sequence connectivity. Making analogy between faces pixels image allows us present convolution operator aggregate local features from nearby By exploiting neighborhoods, can support standard convolutional network concepts, e.g. variable kernel size, stride, dilation. Based hierarchy, we make use pooling layers uniformly merge four into one upsampling method splits four. Thereby, many popular architectures be easily adapted process Meshes arbitrary connectivity remeshed via self-parameterization, making SubdivNet general approach. Extensive evaluation various applications demonstrate SubdivNet's effectiveness efficiency.