Lightning Pose Homepageο
π’ Official Release: Lightning Pose App 2.0
A new modern UI is available for multi-camera single-animal pose estimation, featuring end-to-end support for labeling, model management, and viewing predictions.
Since the initial release in late Jan 2026. Over 70 researchers from 10+ countries have installed and run the app.
Get started: Check out the Create your first project tutorial and installation guide.
See whatβs new: Check out the release notes for the latest updates. New capabilities are released weekly.
π£ Join the discord: Have a question or feature request? Just want to chat? Weβd love to hear from you in the Discord channel!
An end-to-end toolkit for robust multi-view animal pose estimation.
Multiview transformers and patch masking for robust 3D tracking.
Temporal context networks that learn from unlabeled video.
Browser-based labeling and training on headless GPU servers.
Multi-view Capabilitiesο
Multi-View Transformer (MVT): A unified architecture that enables simultaneous processing of information across all camera views through early-feature fusion.
Patch Masking: A novel training scheme that masks random image patches to force the model to learn robust cross-view correspondences.
Geometric Consistency: For calibrated setups, the framework incorporates 3D triangulation losses and geometrically-aware 3D data augmentation.
Variance Inflation: An advanced technique for outlier detection that identifies geometrically inconsistent predictions.
Single-view Capabilitiesο
Temporal Context Networks: Utilizes information from surrounding frames to resolve anatomical ambiguities and maintain tracking through brief occlusions.
Unsupervised Learning: Employs training objectives that penalize predictions for violating physical laws.
Pretrained Backbone Support: Optimized to work with generic, off-the-shelf Vision Transformer (ViT) backbones.
Cloud Application & Workflowο
Cloud & Headless Compatibility: A browser-based interface designed for local or cloud deployment.
Multi-view Labeling: A specialized annotation tool that streamlines the labeling process by using camera calibration.
Unified Multi-view Viewer: Integrated visualization tools to inspect and compare predictions across all camera views simultaneously.
Read the papersο
The original Nature Methods 2024 paper that introduced Lightning Pose for single-view pose estimation using semisupervised learning and ensemble kalman smoothing (EKS):
The 2026 preprint that added robust multiview support using multiview transformers (MVT), patch masking, 3d image augmentation and losses, and multiview EKS.
Get started with the appο
The lightning pose app provides an easy-to-use GUI to access most lightning pose features.
To get started, install lightning pose and follow the Create your first project tutorial. It covers the end-to-end workflow of labeling, training, and evaluation.