MVTEC 3D-AD

该数据集名为MVTec 3D-AD，是首个3D工业异常检测数据集，包含2656个训练样本和1137个测试样本。训练集包含无缺陷的数据，而测试集则包括无缺陷和有缺陷的样本，后者带有像素级标签，指明缺陷位置。每个类别都提供了2D RGB图像和3D点云扫描，其中3D扫描是通过结构光工业传感器获得的。该数据集涵盖10个类别，包括划痕、孔洞、切割、色差等，其任务是进行3D工业异常检测。

The dataset named MVTec 3D-AD is the first dedicated 3D industrial anomaly detection dataset, comprising 2656 training samples and 1137 test samples. The training set exclusively contains defect-free specimens, whereas the test set encompasses both defect-free and defective samples, with the defective ones being annotated with pixel-level labels that indicate the exact locations of defects. For each category, both 2D RGB images and 3D point cloud scans are provided, and the 3D scans are acquired using structured-light industrial sensors. This dataset covers 10 categories including scratches, holes, cuts, color deviations and other common industrial defects, with its core task being 3D industrial anomaly detection.

搜集汇总

背景与挑战

背景概述

MVTec 3D-AD是一个用于无监督3D异常检测和定位的数据集，包含10个类别的4000多个高分辨率扫描样本，提供无缺陷训练样本和带缺陷测试样本，并附有精确的地面真实注释。该数据集适用于学术研究，遵循CC BY-NC-SA 4.0许可，禁止商业用途。