GetMesh: A Controllable Model for High-quality Mesh Generation and Manipulation

TL;DR

This paper proposes a highly controllable generative model, GetMesh, for mesh generation and manipulation across different categories, which generates meshes with rich and sharp details, outperforming both single-category and multi-category counterparts.

Abstract

Mesh is a fundamental representation of 3D assets in various industrial applications, and is widely supported by professional softwares. However, due to its irregular structure, mesh creation and manipulation is often time-consuming and labor-intensive. In this paper, we propose a highly controllable generative model, GetMesh, for mesh generation and manipulation across different categories. By taking a varying number of points as the latent representation, and re-organizing them as triplane representation, GetMesh generates meshes with rich and sharp details, outperforming both single-category and multi-category counterparts. Moreover, it also enables fine-grained control over the generation process that previous mesh generative models cannot achieve, where changing global/local mesh topologies, adding/removing mesh parts, and combining mesh parts across categories can be intuitively, efficiently, and robustly accomplished by adjusting the number, positions or features of latent points. Project page is https://getmesh.github.io.

Figures (24)

• Figure 1: Meshes generated by our method. GetMesh is able to generate diverse and high-quality meshes across the $55$ categories in ShapeNet.
• Figure 2: Overview of the mesh autoencoder. Points are sampled from the surface of the input mesh and encoded to a varying number of latent points. The latent point representation is re-organized to the triplane representation by projecting the points to the triplane. DMTet is utilized to extract a coarse mesh from the triplane and a refinement module further refines the coarse mesh.
• Figure 3: Architecture of the refinement module.
• Figure 4: Visual comparison between meshes generated by our method and baselines. Zoom in to better see the details. More qualitative results are in Appendix Section \ref{['sec: more_qualitative_results']}.
• Figure 5: Compare meshes reconstructed by autoencoders with and without the refinement module. For each pair of meshes, the left one is without the refinement module, and the right one is with the module. Zoom in to see more details.