MVTecAD

该数据集专为工业环境中的异常检测而设计，包含了多种物体及其缺陷。它既包含正常数据也包含明显的异常数据，这使得它非常适合用来评估异常检测方法。该数据集的规模涉及5个物体类别，每个类别都有正常和异常的样本。其任务类型为图像单类分类（I-Occ）。

This dataset is specifically designed for anomaly detection in industrial environments, covering various objects and their defects. It contains both normal samples and distinct anomalous samples, making it highly suitable for evaluating anomaly detection approaches. This dataset encompasses 5 object categories, with normal and anomalous samples available for each category. Its task type is single-class image classification (I-Occ).

搜集汇总

背景与挑战

背景概述

MVTec AD是一个专注于工业检测的异常检测基准数据集，包含超过5000张高分辨率图像，分为15个物体和纹理类别，每个类别提供无缺陷训练图像和带缺陷测试集。该数据集提供所有异常的像素级标注，但仅限非商业用途，采用CC BY-NC-SA 4.0许可证。