Asphalt Pavement Crack Segmentation Dataset for Mobile Road Distress Assessment

This dataset contains asphalt pavement images and corresponding binary segmentation masks for crack and pavement distress detection. It was created to support the development and evaluation of computer-vision and deep-learning algorithms for real-time pavement distress segmentation. The dataset includes image–mask pairs collected from heterogeneous sources and acquisition conditions. The release includes manually collected and manually annotated images acquired using both handheld devices and a vehicle-mounted sensing platform. The images include non-perpendicular viewpoints, variable illumination, shadows, asphalt patches, road markings, curbs, drainage structures, vehicles, manholes, and other objects that an operational model should ignore. The annotations are provided as binary masks, where crack or pavement distress pixels are highlighted against the background. The dataset was used to train and evaluate semantic segmentation models for pavement distress assessment, including DeepLabV3-based architectures deployed on an NVIDIA Jetson edge-computing platform. The current version of the dataset contains 2209 images with associated masks split into 1480 training samples, 498 validation samples, and 231 test samples. The data support research on pavement crack segmentation, road condition monitoring, mobile sensing, smart-city infrastructure assessment, and embedded/edge-AI vision systems.