Grand Bassin Traffic Dataset: Anonymised Aerial and Roadside Traffic Imagery from the Maha Shivaratri Pilgrimage, Mauritius (v1.0)

An anonymised, auto-annotated image dataset of vehicle and pedestrian traffic during the Maha Shivaratri pilgrimage at Grand Bassin (Ganga Talao), Mauritius, captured simultaneously from an overhead near-nadir parking-area camera and a roadside CCTV camera. The paired viewpoints make it suitable for studying the aerial-vs-ground domain gap in object detection. 4,000 JPEG frames (1280x720; 2,000 aerial + 2,000 road) across 24 video segments from 2 cameras, recorded 16-17 February 2023. 854,646 COCO-format bounding boxes across six populated categories, with per-box detector and confidence-score fields: aerial frames annotated by a VisDrone-finetuned YOLOv8 with SAHI sliced inference (765,249 boxes), road frames by COCO-trained YOLO26m (89,397 boxes). Annotations are machine-generated and not human-verified. All frames are anonymised: 45,978 face regions and 32,854 licence-plate regions pixelated using YuNet/Haar detection plus annotation-driven blurring. Bounding-box geometry is unchanged. Important: frames within a segment are 0.5 s apart; split train/val/test by segment, never by random frame. See DATASET_CARD.md inside the archive for full documentation, ethical-use conditions, and limitations. A dataset descriptor paper and a companion methodology paper are in preparation.