ADer

ADer是一个综合性的视觉异常检测基准数据集，由腾讯优图实验室创建。该数据集包含来自工业、医疗和通用领域的多个子数据集，总计11个，用于评估多种视觉异常检测方法。数据集的创建过程涉及对不同领域数据的收集和标准化处理，以确保评估的公正性和全面性。ADer数据集主要应用于工业质量检测、医疗病变检测等领域，旨在通过无监督学习模式识别图像中的异常区域，推动更强大和可泛化的异常检测系统的发展。ADer is a comprehensive benchmark dataset for visual anomaly detection developed by Tencent YouTu Lab. It consists of 11 sub-datasets spanning industrial, medical and general domains, which are used to evaluate various visual anomaly detection methods. The creation of the ADer dataset involves data collection and standardization across different domains, ensuring the fairness and comprehensiveness of the evaluation. The ADer dataset is primarily applied in fields such as industrial quality inspection and medical lesion detection. It aims to identify abnormal regions in images via unsupervised learning paradigms, promoting the development of more powerful and generalizable anomaly detection systems.

搜集汇总

背景与挑战

背景概述

ADer是一个基于PyTorch的开源视觉异常检测工具箱，支持多种流行的AD数据集和方法，默认在MUAD设置下复现了多种AD方法。该工具箱提供了方便的接口来支持新方法的实现，并支持多类/单类训练和测试、可视化等功能。