Toy Flange Dataset

Classification of 3D models is an open research problem. Many architectures have been proposed to classify 3D models. However, most architectures have tackled the ModelNet (Wu et al. 2015) or ShapeNet (Chang et al. 2015) datasets which offer everyday objects. To test the limitations of existing architectures and motivate new architectures, we present a challenging dataset of 3D models of flanges. There are two variants of flanges in this dataset: 1) 4 holes vs 8 holes and 2) 1 to 9 hole flanges. These holes are placed randomly on the flange faces and also in a regular fashion. Not only does this dataset stress test classification ability of existing architectures, but also promotes visualisation and explanation of learned concepts to match with human reasoning. These models are for demonstration purpose only and do not reflect actual products.

三维模型分类是一项尚未完全解决的开放研究课题。目前已有诸多架构被提出用于三维模型分类任务，但现有相关研究大多聚焦于ModelNet（Wu等人，2015）与ShapeNet（Chang等人，2015）这两类涵盖日常物体的数据集。为检验现有架构的性能局限并推动新型架构的研发，我们构建了一个包含法兰盘三维模型的高挑战性数据集。该数据集包含两类法兰盘变体：其一为4孔与8孔法兰盘的分类任务，其二为孔数覆盖1至9的法兰盘分类任务。这些孔既可以随机排布在法兰盘端面，也可按照规则布局进行排列。本数据集不仅可用于对现有架构的分类能力开展压力测试，还能推动对模型所学概念的可视化与可解释性研究，使其契合人类的推理逻辑。本数据集内的所有模型仅用于演示用途，并不代表真实的工业产品。

搜集汇总

背景与挑战

背景概述

该数据集是一个用于3D模型分类研究的挑战性数据集，专注于法兰（flange）的3D模型，包含两个变体：4孔与8孔法兰以及1到9孔法兰，孔洞以随机和规则方式排列，旨在测试现有分类架构的局限性并促进新方法开发。数据集由Karlsruhe Institute of Technology (KIT)和Endress+Hauser的研究人员创建，发布于2021年，主题领域涉及工程和计算机科学，存档大小为589.3 MB，仅用于演示目的，不反映实际产品。