DeepLabCut is the landmark open-source toolbox for markerless pose estimation with deep learning, letting researchers track user-defined body parts in videos of animals — and humans — without physical markers. Born in the Mathis Lab and published alongside influential papers, it has earned about 5,800 stars on GitHub and remains actively developed, with its latest updates in 2026.
DeepLabCut uses transfer learning from powerful pretrained vision backbones, so it can learn to track new body parts from a remarkably small number of hand-labeled frames — often just a few hundred. The workflow is end to end: extract and label frames in a graphical interface, train a network, evaluate it with built-in diagnostics, refine outliers, and then analyze hours of video to produce trajectories of every tracked point. Modern releases add multi-animal pose estimation, 3D tracking via multi-camera calibration, and pretrained “SuperAnimal” foundation models that work across species with little or no labeling. It has become a standard tool in neuroscience, biomechanics, veterinary science and sports research. The project is maintained by an active research community and ships with extensive documentation and example projects spanning mice, primates, birds, insects and more, while spin-off tools for behavior classification build directly on its pose outputs. Its methods papers have been cited thousands of times.
DeepLabCut is free and open source under the LGPL-3.0 license, including its pretrained models.
DeepLabCut is for neuroscientists, ethologists, biomechanics researchers and clinicians who need precise, automated measurement of movement and behavior from ordinary video.