MVTec AD 2 Dataset
收藏资源简介:
MVTec AD 2数据集由德国MVTec软件有限公司和慕尼黑工业大学共同创建,包含8004张高分辨率图像,涵盖8个具有挑战性的工业异常检测场景。该数据集考虑了透明和重叠物体、暗场和背光照明、正常数据中高度可变的物体以及极小的缺陷等工业检测中常见但未被现有数据集涵盖的复杂情况。数据集设计符合无监督学习设置,训练和验证集只包含正常数据,测试集则包含不同照明条件下的正常和异常数据,以评估模型在真实世界分布变化下的鲁棒性。
The MVTec AD 2 dataset was co-created by MVTec Software GmbH of Germany and the Technical University of Munich. It comprises 8004 high-resolution images covering 8 challenging industrial anomaly detection scenarios. This dataset addresses complex situations commonly encountered in industrial inspection but not covered by existing datasets, including transparent and overlapping objects, dark-field and backlighting illumination, highly variable normal samples, and extremely small defects. The dataset is designed for unsupervised learning setups: the training and validation sets only contain normal data, while the test set includes both normal and abnormal data under various lighting conditions, aiming to evaluate the robustness of models against real-world distribution shifts.

- 1The MVTec AD 2 Dataset: Advanced Scenarios for Unsupervised Anomaly Detection德国MVTec软件有限公司,慕尼黑工业大学 · 2025年



