Tracking Pigeons with Neslihan Wittek

Tracking pigeons with DeepLabCut

Neslihan Wittek, PhD student at the Biopsychology department at Ruhr University Bochum, is currently working on behaviour classification of pigeons.

She collected the data in an experiment related to self-recognition, using 2D (and some 3D) DeepLabCut to further investigate the differences in experimental conditions.

(see more on 2D to 3D-DeepLabCut here)

The network was trained using 420 labelled frames, with the main goal of using the tracked coordinate data not only for basic kinetic analysis such as velocity and trajectory, but also for the classification of individual behaviours as shown in the video below.

Above we see a successful 3D tracking (as a fun weekend project), and below polygon-based rendering of points tracked by DeepLabCut of a pigeon performing tail-shake, head-shake and feather preening behaviours:

Further work will expand on this to add an additional level of machine-learning atop the tracked coordinates for classification of pre-defined behaviours as well as potential detection of new and unknown behavioural patterns.

She also is passionate about open source science, and has helped other get up and running with DeepLabCut! She even gave a talk at PyCon Turkey in 2020 on her use of DeepLabCut!

Read more about her adventures with DeepLabCut, Python, and pigeons on her website, and on GitHub!