Fully Convolutional Mesh Autoencoder using Efficient Spatially Varying Kernels
This work addresses the need for efficient and precise 3D mesh processing for tasks like reconstruction and generation, offering a non-template-specific solution that is applicable to various mesh types, though it appears incremental in improving existing neural autoencoder approaches.
The paper tackles the problem of learning latent representations for arbitrary registered meshes, which previous methods struggled with due to template-specific limitations and poor performance on fine-grained deformations. The proposed fully convolutional mesh autoencoder achieves state-of-the-art reconstruction accuracy and higher interpolation capability in latent codes.
Learning latent representations of registered meshes is useful for many 3D tasks. Techniques have recently shifted to neural mesh autoencoders. Although they demonstrate higher precision than traditional methods, they remain unable to capture fine-grained deformations. Furthermore, these methods can only be applied to a template-specific surface mesh, and is not applicable to more general meshes, like tetrahedrons and non-manifold meshes. While more general graph convolution methods can be employed, they lack performance in reconstruction precision and require higher memory usage. In this paper, we propose a non-template-specific fully convolutional mesh autoencoder for arbitrary registered mesh data. It is enabled by our novel convolution and (un)pooling operators learned with globally shared weights and locally varying coefficients which can efficiently capture the spatially varying contents presented by irregular mesh connections. Our model outperforms state-of-the-art methods on reconstruction accuracy. In addition, the latent codes of our network are fully localized thanks to the fully convolutional structure, and thus have much higher interpolation capability than many traditional 3D mesh generation models.