Meta-ABC: A High-complexity Lattice Geometry Dataset for Deep Learning

High-complexity lattice geometries, known as lattice metamaterials, have various applications and needs in industrial design and manufacture. The popularity of metamaterials has led to many research studies, while it has never been systematically studied in computer vision over a collection of 3D models. We introduce Meta-ABC, a large-scale, multi-format dataset of 3D high-complexity lattice geometries. Our dataset is generated based on the ABC dataset, consisting of over 1000 Computer-Aided Design (CAD) models with several different types of lattice structures. Meta-ABC contains data of various formats and different resolutions, supporting user-specified graphics and vision research applications. We thoroughly analyzed our dataset and performed several baseline experiments on Meta-ABC. Compared with existing datasets in Computer Graphics and Compute Vision areas, extensive comparisons are used to highlight our dataset's unique features and properties in terms of complexity, scalability, and flexibility. Part of the dataset can be temporarily accessed at: MetaABC.