Semi-supervised animal action segmentation: freely moving mouse

In this publicly available dataset, a mouse freely moved around an open arena (Sturman et al. 2020, Neuropsychopharmacology). A camera captured a top-down view of the mouse at 25 Hz. Thirteen points were tracked across the tail, body, ears, and nose of the mouse; the paws were not tracked. This repository contains the raw pose data, as well as a set of 21 features computed from the pose data based on distances, angles, and areas. This provides a position- and orientation-invariant representation. We also include the distance between the centroid of the mouse to the boundary of the arena.We consider all three behavior categories provided in the public dataset: unsupported rearing, supported rearing (when the mouse uses the wall of the arena), and grooming. In the rest of the frames, there is no specific action or behavior displayed, so we do not train any of the models to identify this “other” category. We only consider the three specific behaviors described above in our analyses. We use 10 labeled videos for training, and 10 for testing. The dataset contains 280k frames, 83k of which are labeled.Many thanks to the authors of the "Deep learning-based behavioral analysis reaches human accuracy and is capable of outperforming commercial solutions" (Sturman et al. 2020, Neuropsychopharmacology) paper who collected and analyzed the original video dataset: Oliver Sturman, Lukas von Ziegler, Christa Schläppi, Furkan Akyol, Mattia Privitera, Daria Slominski, Christina Grimm, Laetitia Thieren, Valerio Zerbi, Benjamin Grewe & Johannes Bohacek.