Australian Native Bee Detection and Tracking Dataset

This dataset supports research on automated visual monitoring of Australian native bees at hive and nest entrances using computer vision. It comprises four components: 1. Annotated Stingless Bee Detection Dataset 3,277 annotated frames extracted from 16 video clips recorded at a Tetragonula carbonaria (stingless bee) hive entrance (1,280 × 720 px, 25 fps). Each frame includes bounding box annotations in YOLO format (class, normalised centre x, centre y, width, height). Suitable for training and evaluating single-class object detectors. 2. Annotated Stingless Bee Tracking Dataset 16 video sequences (3,511 frames from 13 sequences used for evaluation; 3 short sequences < 20 frames included for completeness) with per-frame bounding boxes and unique bee identity labels in YOLO format (bee_N class IDs). Supports multi-object tracking evaluation using MOT Challenge–compatible ground truth. Collection details: All footage was recorded in 2024 at hive/nest entrances in Victoria, Australia. Video resolution is 1,280 × 720 pixels at 25 fps. Annotations were produced manually using YOLO-format labelling. Associated code: Training scripts, evaluation pipelines, and pre-computed benchmark results are available at https://github.com/asadiceiu/native-bee-detection-tracking. Potential uses: Training and benchmarking object detectors and multi-object trackers; ecological studies of foraging activity and hive traffic; cross-species comparison of visual monitoring methods for native bees.