IRD-dataset: a multi-source Iraqi road defect dataset for detection, segmentation, and GPS-based mapping

Road defect datasets support the development of computer vision methods for pavement inspection, road condition assessment, and maintenance planning. The public pavement defect datasets cited in this article are based on data collected outside Iraq. This article introduces IRD-Dataset, a public multi-source dataset documenting paved road environments in Baghdad, Iraq. The dataset consists of 4,352 RGB images collected from public paved road environments using three sources: Dashcam (1,006 images), Drone (2,162 images), and Mobile (1,184 images). Images were captured under varied field conditions and screened to retain those suitable for annotation and public release. Five road surface categories are included in the dataset: longitudinal crack, transverse crack, alligator crack, pothole, and speed bump. The first four categories are pavement defects, whereas speed bumps are intentional road features included for practical road-scene analysis. The images were manually annotated with bounding boxes for object detection and polygons for instance segmentation. The release also provides segmentation masks, image-level GPS metadata, 401 background images, and a YOLO configuration file. IRD-Dataset supports annotation-derived image classification, object detection, instance segmentation, semantic segmentation mask generation, and image-level GPS mapping of road surface observations. The data may support pavement monitoring, municipal road maintenance planning, and smart city road condition analysis.