PKU-GoodsAD
收藏资源简介:
PKU-GoodsAD数据集是专为无监督异常检测设计的,包含6124张超市商品的高分辨率图像,涵盖484种不同商品,分为6大类。每类商品包含多种异常类型,如变形、表面损伤和开口等。数据集遵循无监督设置,仅使用无缺陷图像进行训练,提供像素级精确的异常区域标注。该数据集旨在推动无人工超市和智能制造领域的异常检测研究和应用,通过评估现有无监督异常检测方法,揭示了不同方法在实际应用中的性能差异。
The PKU-GoodsAD dataset is specifically designed for unsupervised anomaly detection. It contains 6,124 high-resolution images of supermarket merchandise, encompassing 484 distinct product types grouped into 6 major categories. Each category features multiple anomaly types, including deformation, surface damage, and opening defects. The dataset adheres to the unsupervised learning paradigm: only defect-free images are utilized for model training, with pixel-perfect annotations for anomalous regions provided. This dataset is intended to advance anomaly detection research and applications in the domains of unmanned supermarkets and intelligent manufacturing. By evaluating existing unsupervised anomaly detection methods, it exposes the performance gaps between different methods in real-world scenarios.




