Australian Native Bee Detection and Tracking Dataset with 3,277 Annotated Frames

3,277 annotated frames and 16 video sequences support automated visual monitoring of Australian native bees. The dataset, created by Asad Asaduzzaman in 2024, includes bounding box annotations for detection and unique identity labels for multi-object tracking. Associated training scripts and evaluation pipelines are available on GitHub.

Use Cases

• Training single-class object detectors based on annotated bounding boxes for stingless bees.
• Evaluating multi-object tracking algorithms based on MOT Challenge–compatible ground truth with unique bee IDs.
• Conducting ecological studies of hive traffic and foraging activity based on video sequences recorded at hive entrances.
• Comparing visual monitoring methods for native bees across different species based on the provided benchmark results.

Strengths

• Includes 3,277 manually annotated frames for object detection.
• Provides 16 video sequences with per-frame bounding boxes and unique identity labels for tracking evaluation.
• Associated code, including training scripts and evaluation pipelines, is publicly available on GitHub.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic bias inherent to its collection in Victoria, Australia.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Asad Asaduzzaman

Collection Method

Footage recorded at hive/nest entrances; annotations produced manually using YOLO-format labelling.

Time Range

2024

Freshness

Last updated 2026-05-27 04:17:51; freshness should be verified.

Geography

Victoria, Australia