2026-10-07T22:45:48Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33102095 2026-09-13T19:15:58Z category_26377 category_28879 item_type_3 month_year_09_2026

NSVD: Vehicle Image Dataset from Northern Slovakia under Diverse Weather Conditions University Science Park UNIZA (UVP), University of Zilina (24472823) Transport engineering Computer vision vehicle detection object detection YOLO roadside camera adverse weather weather labels traffic monitoring van class benchmark dataset Slovakia computer vision <p>Created by the University Science Park of the University of Zilina (UVP UNIZA), Zilina, Slovakia.</p> <p>We kindly ask you to cite the following manuscript:</p> <p>[Northern Slovakia Vehicle Dataset (NSVD) manuscript – title, venue and DOI to be added upon publication].</p> <p>Thank you!</p> <p>The dataset contains:</p> <ul> <li>8000 human-verified roadside images in Full HD resolution (1920 x 1088), about 2000 for each of four weather conditions: sunny, cloudy, rain and snow;</li> <li>70,772 checked bounding boxes in YOLO format (normalised: class x_center y_center width height), for five declared classes, 0 car, 1 bus, 2 truck, 3 van, 4 motorcycle, with counts of 36,877 car, 6,551 bus, 22,577 truck and 4,767 van. The motorcycle class is reserved and empty, because recording ran outside the season of powered two-wheelers. The native van class covers van-bodied vehicles that the MS COCO taxonomy cannot express;</li> <li>the weather condition and the recording site of every image, in three forms: a metadata table, COCO files in which each image record carries site and weather, and image lists grouped by site, by weather and by both;</li> <li>the predefined split of 5527 training, 1128 validation and 1199 test images, and leave-one-site-out folds that hold each site out of training entirely;</li> <li>ground truth for two evaluation protocols, the three classes shared with MS COCO and the full four-class taxonomy, with a short example script for scoring predictions;</li> <li>documentation, the recording geometry, a per-image MD5 list and the licence.</li> </ul> <p>Images were recorded by one camera moved between four sites in the Zilina Region of northern Slovakia, mounted about 4.5 m above the road at a vertical angle of 27 degrees. Three sites lie on first-class roads I/11 and I/18, the fourth in the centre of Zilina. Faces and licence plates are blurred. The image files are identical to those of the previous release.</p> 2026-09-13T19:15:58Z Dataset Dataset 10.6084/m9.figshare.33102095.v3 https://figshare.com/articles/dataset/CERVD_A_Human-Verified_Weather-Balanced_Slovak_Roadside_Vehicle-Detection_Dataset_with_YOLO_Annotations/33102095 CC BY 4.0