MarmoPose: A Deep Learning-Based System for Real-time Multi-Marmoset 3D Pose Tracking

Chaoqun Cheng,Zijian Huang,Ruiming Zhang,Guozheng Huang,Han Wang,Likai Tang,Xiaoqin Wang

DOI: https://doi.org/10.1101/2024.02.27.582429

2024-01-01

Abstract:The common marmoset has become an important experimental animal model in scientific research. The ability to capture and quantify behaviors of marmosets in natural environment and social scenarios is highly desired by marmoset research community. Although existing pose tracking methods have enabled multi-marmoset two-dimensional (2D) tracking and single-marmoset three-dimensional (3D) pose estimation, they have not fully addressed the practical demands of marmoset research. Here, we introduce MarmoPose, a real-time 3D pose tracking system based on deep learning for automatically estimating 3D poses of multiple marmosets freely roaming in their homecage. MarmoPose employs a marmoset skeleton model to optimize 3D pose computation and estimate invisible body locations. Furthermore, MarmoPose provides a real-time processing module to enable short-latency closed-loop experimental control based on the 3D poses of marmosets. The objective of our effort is to design an efficient, cost-effective and user-friendly system that can be easily adapted by marmoset researchers to quantify natural behaviors in typical housing environment. ### Competing Interest Statement The authors have declared no competing interest.