UAV Road Defect Segmentation Dataset (RDD2022-derived)

This dataset provides pixel-level ground-truth annotations for crack-like road surface defects, derived from the publicly available Road Damage Detection 2022 (RDD2022) dataset. While RDD2022 provides bounding-box annotations suitable for object detection, this dataset introduces manually annotated pixel-level segmentation masks to accurately capture defect geometry. A total of 400 images were manually annotated at pixel level. RGB images are provided in JPG format, while the corresponding binary segmentation masks are provided in PNG format with matching filenames. The dataset is split into training, validation, and test sets following an 80/15/5 ratio (320/60/20), using a fixed random seed to ensure reproducibility. The dataset is intended to support research on semantic segmentation, contour extraction, and downstream geospatial road inspection workflows using UAV imagery.