This paper introduces a multi-animal pose estimation and tracking system, extending DeepLabCut. It details the development of powerful CNNs (DLCRNet) for animal pose estimation, a data-driven method for automatic skeleton selection, and robust trackers for long-timescale analysis. The system handles multiple animals, even those with similar appearances, and incorporates both supervised and unsupervised animal identity tracking to enhance performance. The authors provide open-source benchmark datasets and demonstrate the system's application to analyzing complex social interactions in marmosets.

J. Lauer, M.Z. Alam, S. Yang, S. Schneider, T. Nath, W. Ma, G. Feng, M.M. Reimer, C. Dulac, D. Schneider, V.N. Murthy, V. Di Santo, G. Lauder, A. Mathis, M.W. Mathis