ToyCity

该数据集名为ToyCity，是一个专为场景异常检测（Scene AD）设计的基准数据集。它包含了玩具城市环境中多个真实物体的实拍图像，特点包括建筑物、标志、道路、树木以及其他各类杂物。数据集还包括了从版本v1到v4的不同增强程度和场景布局的多个版本，并且为异常分割提供了手动标注的地面真实数据。与单一物体数据集如MAD相比，该数据集含有数量多得多的真实图像。其任务是场景异常检测。

The dataset, named ToyCity, is a benchmark dataset designed for scene anomaly detection (Scene AD). It contains real captured images of multiple real objects in a toy city environment, featuring buildings, signs, roads, trees, and various other clutter. The dataset also includes multiple versions ranging from v1 to v4 with varying degrees of enhancement and scene layouts, and provides manually annotated ground-truth data for anomaly segmentation. Compared with single-object datasets such as MAD, this dataset contains a significantly larger number of real images. Its task is scene anomaly detection.