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

• DOI: 10.48550/arxiv.2403.11990
• Publication Date: 2024-03-18
• Authors: Zhaoyang Lyu; Ben Fei; Jinyi Wang; Xudong Xu; Ya Zhang; Weidong Yang; Bo Dai
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

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