Urban-Anomaly Dataset: A Multi-Source Benchmark for Multi-Class Urban Anomaly Detection

Urban-Anomaly is a multi-source dataset developed for multi-class urban anomaly detection. The dataset integrates publicly available resources and supplementary urban images collected from different scenarios. The dataset contains seven representative categories of urban anomalies, including RoadCracks, Potholes, DamagedRoadSigns, FallenTrees, Garbage, Graffiti, and DamagedElectricalPoles. The dataset was constructed through data collection, selection, cleaning, duplicate removal, annotation verification, and category reorganization. It is designed to support research on real-time object detection methods for complex urban environments. The dataset is associated with the manuscript:"UAD-YOLO: An Efficient Context-Aware and Shape-Aware Framework Based on YOLOv11 for Urban Anomaly Detection in Complex Scenes".