Abstract: Mosquitonet is an open-source pavement distress dataset collected with a low-cost vehicle-mounted system for pavement distress detection through Deep Learning algorithms. There are 7099 images with 13 distress classes, annotated by experts in 12 formats: COCO, PASCAL VOC, YOLO/YOLOv4-v7 PyTorch, TF Object Detection, TFRecord, YOLOv3 Keras, Retinanet Keras and CreateML. The images were gathered by means of a low-cost, high-resolution and fast-acquisition system, Mosquito. The Mosquito system is an unmanned aerial vehicle mounted on a 3D-printed structure that is attached to a vehicle using a suction cup system. The Mosquito system provides images and GPS coordinates per image for speeds up to 120 km/h, where its remote control allows the co-pilot to adjust its parameters in real time for better capture. Annotations for training Deep Learning models were performed manually by pavement experts.
