Lightning Pose
Open-source animal pose estimation software for 2D and 3D tracking. Runs on Linux or WSL with an NVIDIA GPU and includes a labeling and evaluation GUI.
Lightning Pose is an open-source package for researchers who estimate animal poses from video for behavior analysis. It supports 2D and 3D pose estimation with single-camera and multiple-camera datasets, including footage where body parts become hidden from view.
Its multi-view models use transformer architectures to learn geometric relationships between camera views. Training with simulated occlusions helps the models handle obscured body parts. For single-view footage, the models use information across time and unsupervised losses to improve estimates in difficult scenes. This semi-supervised approach combines labeled examples with learning signals that don't require manual labels.
The GUI covers labeling, model management, and evaluation, so researchers can work through those stages within the same application. The package also provides a command-line interface and a Python API for workflows that need scripting.
Lightning Pose runs locally on Linux or WSL and requires an NVIDIA GPU. Training time depends heavily on the dataset size and available GPU resources. Researchers without a local NVIDIA GPU can use the Lightning AI cloud environment instead, where the workload runs on cloud hardware. The Python package uses the MIT license.
