Context We use honeybees as a system to develop dense object detection and tracking algorithms. Human crowds, cells in a tissue, animal groups are all composed of densely packed highly similar objects in motion. Honeybees live on a 2D surface of the honey comb and are an easily accessible for imaging. Their movement is irregular and sometimes very rapid. Resolving tracking in such environment could help finding solutions for tracking in other multi-object systems. Content The dataset contains video frames of 2 recordings (recorded at 30fps and 70fps and named so) downsampled to 2fps. There is 360 frames of each recording, in addition the dataset contains text files with positions and orientation angles of bees in those frames images. The marked bees belong to selected regions with high bee density within 1024x1024 windows [as we describe here][1], the offsets of these windows are indicated in the data. There are 2 object classes: (1) fully visible bees, (2) abdomens of bees hidden inside cells of a honey comb. The orientation angle of objects of class 2 is always 0. The format of text files is: offset_x offset_y class position_x position_y angle Acknowledgements This data was collected with the use of the generous funding of the [Okinawa Institute of Science and Technology][2]. We would also like thank all mechanical turk workers that provided great work to make our research possible. Inspiration Can we robustly detect bees in the images? How can we track the highly similar and dense individuals? Can ML help this task that has been so far intractable for computer vision? [1]: https://groups.oist.jp/bptu/honeybee-tracking-dataset [2]: http://oist.jp