About
SLEAP (Social LEAP Estimates Animal Poses) is a powerful open-source tool designed for researchers in behavioral neuroscience and animal behavior science. It enables users to train and deploy deep learning models that automatically track any number or type of animal across recorded video footage with high precision and speed. The platform features a purpose-built graphical interface with an efficient human-in-the-loop labeling workflow, allowing researchers to rapidly create annotated datasets and train accurate pose estimation models with very few labels. SLEAP supports both single and multi-animal scenarios using top-down and bottom-up training strategies, making it adaptable to a wide range of experimental setups. With the release of SLEAP 1.5+, the neural network backend was migrated from TensorFlow to PyTorch, delivering dramatically faster training and inference. The system achieves over 800 FPS real-time tracking throughput and sub-3.5ms per-frame latency, enabling live experimental feedback loops. Two standalone libraries accompany the GUI: sleap-io for reading, writing, and manipulating .slp data files, and sleap-nn for headless PyTorch-based training and inference pipelines suited to remote servers and custom workflows. SLEAP has been adopted by 33,000+ users, downloaded 150,000+ times, and cited in 1,500+ scientific publications. It runs on all major operating systems (Windows, macOS, Linux) with full CUDA GPU support and is installable via uv or pip.
Key Features
- Purpose-Built Labeling GUI: Advanced graphical interface with a human-in-the-loop workflow for rapid, efficient creation of annotated pose estimation datasets.
- Multi-Animal Pose Estimation: Supports single and multi-animal tracking using top-down and bottom-up deep learning strategies to handle diverse experimental setups.
- High-Speed PyTorch Inference: Achieves 800+ FPS real-time tracking throughput and <3.5ms per-frame latency using a modern PyTorch backend, enabling live experiment feedback.
- Standalone Python Libraries: Includes sleap-io for .slp file handling and sleap-nn for headless server training/inference, making SLEAP fully modular for custom pipelines.
- Customizable Model Architectures: Flexible neural network architectures that deliver accurate predictions even with very few labeled examples, reducing annotation burden.
Use Cases
- Tracking the body pose of multiple mice simultaneously in open-field behavioral experiments to quantify locomotion and social interaction.
- Analyzing insect movement in laboratory arenas to study neural correlates of navigation and decision-making.
- Building real-time closed-loop neuroscience experiments that trigger stimuli based on an animal's live detected pose.
- Generating large annotated pose datasets from video recordings for downstream behavioral classification and machine learning research.
- Processing video from remote or server-based experimental rigs using the headless sleap-nn pipeline without the graphical interface.
Pros
- Completely Free and Open Source: SLEAP is fully open source with no licensing fees, making it accessible to academic labs and independent researchers worldwide.
- Exceptional Performance: With 800+ FPS inference and sub-10ms latency, SLEAP is fast enough for real-time tracking in closed-loop behavioral experiments.
- Broad Platform & GPU Support: Runs on Windows, macOS, and Linux with full CUDA support, and installs easily via uv or pip for quick onboarding.
- Modular Architecture: Standalone sleap-io and sleap-nn packages let researchers use specific components independently, enabling flexible integration into custom pipelines.
Cons
- Steep Learning Curve for Non-Developers: While the GUI is helpful, getting the most out of SLEAP — especially custom training pipelines — requires comfort with Python and command-line tools.
- GPU Required for Practical Training: CPU-only installation is available but model training and high-speed inference practically require a CUDA-compatible NVIDIA GPU.
- Breaking Changes in v1.5+: The migration from TensorFlow to PyTorch in v1.5 introduced significant breaking changes that may require workflow updates for existing users.
Frequently Asked Questions
What is SLEAP used for?
SLEAP is used to automatically track the body pose of one or multiple animals in recorded videos using deep learning. It is primarily used in behavioral neuroscience and animal behavior research to quantify movement and behavior with high precision.
Is SLEAP free to use?
Yes, SLEAP is completely free and open-source. It is available on GitHub and installable via uv or pip with no licensing restrictions.
What operating systems does SLEAP support?
SLEAP supports Windows, macOS, and Linux. GPU-accelerated training and inference require a CUDA-compatible NVIDIA GPU, but a CPU-only install is available for inference and labeling.
What are sleap-io and sleap-nn?
sleap-io is a standalone Python library for reading, writing, and manipulating SLEAP's .slp data files. sleap-nn is a PyTorch-based library for training and running pose estimation models, suitable for headless server use without the GUI.
How fast is SLEAP's inference?
SLEAP can process over 800 frames per second in batch inference mode with less than 3.5ms per-frame latency, making it suitable for real-time closed-loop experimental setups.
