MAD-Bench
收藏资源简介:
MAD-Bench是一个用于多层次异常检测的基准数据集,由新加坡国立大学和新加坡科技设计大学等机构创建。该数据集包含31个子集,涵盖了新颖性检测、工业检测和医学影像等多个领域。数据集通过手动分配严重性级别,确保模型能够准确评估异常的严重程度。创建过程结合了现有的多种数据集,并进行了适当的调整和标注。MAD-Bench旨在解决现有模型在异常检测中无法准确反映实际严重程度的问题,适用于需要精细区分异常严重性的应用场景。
MAD-Bench is a benchmark dataset for multi-level anomaly detection, developed by institutions including the National University of Singapore and the Singapore University of Technology and Design. This dataset comprises 31 subsets spanning multiple domains such as novelty detection, industrial inspection, and medical imaging. The severity levels of anomalies are manually assigned to ensure that models can accurately evaluate the severity of anomalies. Its development process combines multiple existing datasets with appropriate adjustments and annotations. MAD-Bench aims to address the issue that existing anomaly detection models fail to accurately reflect the actual severity of anomalies, and is suitable for application scenarios requiring fine-grained differentiation of anomaly severity.




