Primate Markerless Pose Estimation and Movement Analysis Using DeepLabCut
Rollyn T. Labuguen, Dean Karlo Bardeloza, Salvador Blanco Negrete, Jumpei Matsumoto, Ken‐ichi Inoue, Tomohiro Shibata · 2019
The analysis and understanding of primate behavior play a fundamental role in fields such as neuroscience, medicine, psychology, genetics, and more. This paper demonstrates an automatic detection of primate features by using an open-source deep learning toolset, DeepLabCut. We trained the deep neural network to locate 17 features and extract the monkey pose by relating the set of features detected. The model is trained with 5,967 manually annotated monkey images, which achieved train and test set errors of 3.61 and 19.72 unit pixels respectively. We also plotted the feature trajectories across multiple frames to show that the trained model can be used for analyzing behavior.