UVY: a Dataset of Sport User-Generated Videos from YouTube for Multi-Object Tracking Evaluation

UVY (Sport User-generated Videos from YouTube), a dataset of annotated sports videos collected from YouTube to support the adaptation, fine-tuning, and evaluation of Multi-Object Tracking (MOT) models in an unconstrained, user-generated videos (UGV) context. The dataset contains 20 UGVs covering seven sports categories: soccer, basketball, volleyball, tennis, athletics, artistic gymnastics, and cricket. It contains a total of 31,961 annotated frames, 128,258 bounding boxes, and 378 annotated tracks. Videos were collected under Creative Commons (CC BY) licenses and annotated using a semi-automated pipeline that combines YOLO-World (yolo_world/s) zero-shot object detection with manual review and correction via CVAT (Computer Vision Annotation Tool).