Open-Industry
收藏资源简介:
Open-Industry是由多所高校联合构建的工业级3D点云异常检测数据集,包含15类常见工业部件共计2400个样本。该数据集采集自真实生产线,每类包含100个正常样本和60个异常样本,涵盖凸起、凹陷、划痕等五种典型缺陷类型,平均异常点比例仅1.2%。通过结构化光采集系统获取多视角点云后,经过校准、配准和降噪处理,最终形成单样本约9万点的标准化数据。该数据集专为开放集监督的3D异常检测任务设计,旨在解决工业质检中真实缺陷样本稀缺条件下的未知异常识别难题。
Open-Industry is an industrial-grade 3D point cloud anomaly detection dataset jointly constructed by multiple universities, comprising a total of 2400 samples across 15 categories of common industrial components. This dataset is collected from real production lines, with each category containing 100 normal samples and 60 abnormal samples, covering five typical defect types such as bumps, dents, scratches and others, with an average proportion of abnormal points being only 1.2%. After acquiring multi-view point clouds via structured light acquisition systems, the dataset undergoes calibration, registration and denoising processing, ultimately forming standardized data with approximately 90,000 points per sample. This dataset is specifically designed for open-set supervised 3D anomaly detection tasks, aiming to address the challenge of unknown anomaly recognition under the condition of scarce real defect samples in industrial quality inspection.
OpenIndustry 数据集概述
数据集基本信息
- 数据集名称:OpenIndustry
- 核心研究领域:开放集监督式三维异常检测
- 主要应用场景:工业场景,专注于检测训练阶段未观察到的未知缺陷
数据集内容与目标
- 旨在支持开放集监督式三维异常检测的研究
- 当前发布阶段提供适用于自监督异常检测任务的数据集划分
- 未来版本将通过官方数据加载器自动生成论文中介绍的开放集设定划分
数据获取与发布状态
- 数据集链接:https://huggingface.co/datasets/HanzheL/open-industry/upload/main
- 当前发布状态:已完成步骤一(OpenIndustry数据集,自监督划分)
- 待发布内容:
- 步骤二:论文中使用的基准实现
- 步骤三:Open3D-AD框架(用于开放集监督式三维异常检测的可泛化框架)
相关论文
- 论文标题:Open-Set Supervised 3D Anomaly Detection: An Industrial Dataset and a Generalisable Framework for Unknown Defects
- arXiv链接:https://arxiv.org/abs/2604.01171
- arXiv ID:2604.01171
- 发表年份:2026




