PaveDistress: A comprehensive dataset of pavement distresses detection

The PaveDistress dataset is a high-quality collection of road surface images designed for non-destructive detection of pavement distresses like cracks, patches, and potholes. Captured using a specialized inspection vehicle, the images were taken along the S315 highway in China, covering diverse lighting conditions and real-world scenarios such as shadows, oil stains, and debris. Each image has a resolution of 3854 × 2065 pixels, representing an actual road coverage area of 3.9m × 2.1m. The dataset is organized into categories including cracks (transverse, longitudinal, and map cracks), patches, potholes, and background images without defects. This dataset supports the development of deep learning models for detecting, classifying, and segmenting road distresses, contributing to improved automation in pavement maintenance systems. Its diverse and representative content makes it valuable for research in civil engineering, enabling machine learning applications for road monitoring and condition assessment. Researchers can use this dataset to train models for accurate detection, which can be applied in real-world road maintenance operations.