CoCG Road Condition - Oriented Bounding Boxes (CoCGRCOBB)

CoCG Road Condition - Oriented Bounding Boxes (CoCGRCOBB) is an extension of the Camera on Car Grille Road Condition - Detection Dataset (CoCGRCDD) using Oriented Bounding Boxes annotation. Research conducted on the CoCGRCDD dataset showed that the object detection models tested did not generalize transverse and longitudinal cracks between each other very well, and it was therefore decided to use Oriented Bounding Boxes to combine the two classes C01 and C02 into a single class C12. This data set contains frames from video recordings of road pavements in Poland. The data was obtained from a USB Logitech Brio camera placed on the radiator grille of the car. The annotation process was carried out using CVAT software. Annotations in the form of classes and oriented bounding boxes was saved in MS COCO format (OBB_annotations.json). The dataset contains the following types of road pavement defects: • C12 - single crack and pronounced discontinuity of the material structure in any orientation; • C03 - Alligator cracks and delamination of the surface layer occurring in their area; • C04 - holes on the road surface and larger cavities erosion (such as in the area of cracks). The set consists of 2110 frames from the footage, of which 325 have no road surface defects, which allows us to check the occurrence of the model's prediction quality based on the occurrence of false positives. Each frame in the dataset is additionally annotated (annotations/additional_info.json) with the occurrence of: shadow, painting, outlandish (object that should not be on the road e.g. sand, leaves, etc.), path milling, grain or binder defects, manhole. * The dataset presented here contains only annotations for images that are included in the CoCG Road Condition - Detection Dataset (CoCGRCDD) - https://data.mendeley.com/datasets/snyyfknw56/1.