DeepLabCut

Why we included this project

DeepLabCut is one of the most mature end-to-end toolboxes for turning video of animals or people into per-frame positions of specific body parts. It covers the whole workflow in one package: you label a small set of frames in its graphical interface, train a deep network on those labels, and then run the model on new videos. The pose outputs feed directly into downstream behavior analysis, which is why ethology, neuroscience, and biomechanics labs use it for quantitative tracking without attaching physical markers. Newcomers can fine-tune a pretrained detector from the model zoo instead of training from scratch, and the active community helps when you get stuck. It also works well as a reference if you want to build your own keypoint-estimation pipeline on your own data.