milimlee-synth3d/objaverse.data.intelligence

Objaverse Data Intelligence 数据集提供了对约68万个Objaverse资产的大规模场景级几何、结构和材料分析，旨在为机器学习提供过滤、结构分解和数据集管理支持。该数据集从复杂3D场景中提取结构化元数据，涵盖几何、拓扑、场景结构、材料属性和分解特性。它支持检测常见场景特征（如动画资产、大型平面、环境球体/立方体、隐藏内部几何、原始扫描资产、水密网格等），分析材料属性（如颜色、纹理、透明度、默认材料），并提供结构分解元数据以识别多组件场景。数据集旨在帮助研究人员和工程师在训练或评估前理解场景组成和数据集质量，侧重于场景级结构理解和分解指导，而非语义对象分类。Objaverse Data Intelligence provides large-scale scene-level geometry, structure, and material analysis of ~680K Objaverse assets, designed for ML-ready filtering, structural decomposition, and dataset curation. It extracts structured metadata from complex 3D scenes, covering geometry, topology, scene structure, material properties, and decomposition characteristics. The dataset enables detection of common scene features (e.g., animation assets, large planes, environment spheres/cubes, hidden interior geometry, raw scanned assets, watertight meshes), analysis of material properties (e.g., color, texture, transparency, default materials), and provides structural decomposition metadata to identify multi-component scenes. It is designed to help researchers and engineers understand scene composition and dataset quality before training or evaluation, focusing on scene-level structural understanding and decomposition guidance rather than semantic object classification.