Dataset capturedbyrealtimevehicle-mountedcamerasystem,  600 high-quality images was extracted, 480 as training set, 120 as valid set. The images have a resolution of 1600x1200 and encompass three types of pavement defects, that is, cracks, patches and potholes. Our dataset is in YOLO format, YOLO (You Only Look Once) is a popular object detection framework that uses a single neural network to predict bounding boxes and class probabilities for various objects in an image. The YOLO dataset format typically consists of two main components: the image files and the annotation files.