ShapeNetCore v2

ShapeNetCore v2，是由斯坦福大学、普林斯顿大学等机构联合发布的一个大规模三维CAD模型数据集，在三维深度学习领域具有重要的基础地位 。ShapeNetCore v2于2017年正式上线，是ShapeNet数据集的一个清洗后且维护良好的子集 。该数据集旨在解决三维数据缺乏统一标准和高质量标注的问题，模型主要来源于3D Warehouse和Yobi3D等公共CAD共享平台 。它是目前三维形状理解（3D Shape Understanding）和重建领域中应用最广泛的标准数据集之一，主要用于支持3D物体分类、三维重建、姿态估计、形状生成等任务的研究与评估 。相比于v1版本，v2在几何结构的完整性、法线朝向的一致性以及类别标注的准确性上都经过了更严格的人工筛选与处理 。该数据集包含经过统一处理的三维CAD模型。为了适应深度学习算法的输入要求，研究人员对原始模型进行了归一化缩放、中心对齐、朝向标准化以及类别映射等预处理操作 。 数据主要以Wavefront OBJ (.obj) 网格格式存储，包含顶点、面片等几何信息，并附带了相应的材质文件（.mtl）和纹理贴图（.png），支持带有纹理的渲染任务 。此外，数据集还提供了丰富的元数据（metadata），包括模型的包围盒（bounding box）、体积、对称性信息以及对齐变换矩阵，均以JSON格式存储 。ShapeNetCore v2涵盖了55个常见的物体类别，具体包括交通工具（如飞机、汽车）、家具（如椅子、桌子）、电子设备、生活用品等 。在数据体量方面，该数据集共收录了51300多个独立的三维CAD模型 。文件组织结构基于语义层级设计，采用WordNet synset编码作为类别目录名，每个模型拥有唯一的ID，确保了数据的组织层次清晰且易于通过语义标签进行索引 。

ShapeNetCore v2 is a large-scale 3D CAD model dataset jointly released by Stanford University, Princeton University and other institutions, which occupies a fundamental foundational position in the field of 3D deep learning. Officially launched in 2017, it is a cleaned and well-maintained subset of the original ShapeNet dataset.

This dataset was developed to address the long-standing issues of lacking unified standards and high-quality annotations for 3D data, with its models primarily sourced from public CAD sharing platforms such as 3D Warehouse and Yobi3D. It is currently one of the most widely adopted standard datasets in the domains of 3D shape understanding and reconstruction, and is extensively used to support research and evaluation for tasks including 3D object classification, 3D reconstruction, pose estimation, and shape generation.

Compared with its v1 iteration, ShapeNetCore v2 has undergone stricter manual screening and processing across three key aspects: geometric structure integrity, consistency of normal orientations, and accuracy of category annotations.

The dataset comprises uniformly processed 3D CAD models. To meet the input requirements of deep learning algorithms, researchers conducted a series of preprocessing operations on the original models, including normalized scaling, center alignment, orientation standardization, and category mapping.

The data is primarily stored in the Wavefront OBJ (.obj) mesh format, which encapsulates geometric information such as vertices and faces, and is accompanied by corresponding material files (.mtl) and texture maps (.png), enabling support for texture-based rendering tasks.

Additionally, the dataset provides comprehensive metadata, including the bounding box, volume, symmetry information of each model, as well as the alignment transformation matrix, all stored in JSON format.

ShapeNetCore v2 covers 55 common object categories, including transportation vehicles (e.g., airplanes, automobiles), furniture (e.g., chairs, tables), electronic devices, daily necessities, and more.

In terms of data scale, the dataset contains over 51,300 independent 3D CAD models. Its file organization follows a semantic hierarchy, using WordNet synset codes as the names of category directories, and each model is assigned a unique ID, ensuring a clear hierarchical structure for data organization and facilitating indexing via semantic tags.

搜集汇总

背景与挑战

背景概述

ShapeNetCore v2是由斯坦福大学、普林斯顿大学等机构联合发布的大规模三维CAD模型数据集，主要用于三维深度学习研究。该数据集包含55个常见物体类别的51300多个三维模型，经过严格筛选和预处理，支持3D物体分类、重建等任务。数据以OBJ格式存储，附带材质和纹理贴图，并提供丰富的元数据。