IRD-Dataset

IRD-Dataset is a public anonymized multi-source Iraqi paved road surface dataset collected in Baghdad, Iraq using Dashcam, Drone, and Mobile cameras. The dataset contains 4,352 RGB road images with YOLO bounding box annotations, YOLO polygon annotations, SegmentationClass masks, SegmentationObject masks, image-level GPS metadata, and a YOLO-compatible configuration file. The dataset includes four road defect classes and one intentional road surface feature. It supports object detection, annotation-derived image-level classification, instance segmentation, semantic segmentation, multi-source road scene analysis, and GPS-based road surface mapping.