Primitive3D

Primitive3D是一个由新加坡国立大学等机构创建的大型3D对象数据集，通过随机组装基本几何体（如球体、立方体等）自动生成，包含150,000个带注释的3D对象。数据集的创建过程利用了构造实体几何（CSG）方案，通过随机化树结构和参数来生成多样化的3D对象。Primitive3D数据集旨在解决3D计算机视觉中高质量数据集的缺乏问题，特别是在大规模和多样性方面。该数据集的应用领域包括3D对象分类、分割和重建等任务，通过多任务学习和数据集蒸馏方法，能够有效提升深度学习模型的性能。

Primitive3D is a large-scale 3D object dataset created by institutions such as the National University of Singapore and other organizations. It is automatically generated via random assembly of basic geometric primitives (e.g., spheres, cubes, etc.) and contains 150,000 annotated 3D objects. The dataset's development process adopts the Constructive Solid Geometry (CSG) framework, generating diverse 3D objects by randomizing the tree structures and their associated parameters. The Primitive3D dataset aims to address the shortage of high-quality datasets in 3D computer vision, particularly in terms of scale and diversity. Its applicable domains cover tasks including 3D object classification, segmentation, reconstruction, and more, and it can effectively improve the performance of deep learning models through multi-task learning and dataset distillation methods.