Openpose
Real-time multi-person keypoint detection library for body, face, hands, and foot estimation.
What is Openpose?
OpenPose is an open-source library for real-time multi-person 2D and 3D keypoint detection, estimating body, face, hand, and foot keypoints from images and video. It solves the problem of accurately and efficiently detecting human pose in complex, multi-person scenarios, enabling applications in motion analysis, human-computer interaction, and augmented reality.
Researchers, developers, and engineers working in computer vision, robotics, human-computer interaction, and augmented reality who need robust, real-time human pose estimation.
Developers choose OpenPose for its real-time performance, multi-person detection with constant runtime, and comprehensive keypoint coverage (135 keypoints across body, face, hands, and feet). Its cross-platform support and flexible APIs make it adaptable for both research and production use.
Overview
OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation
Use Cases
Best For
- Real-time human pose estimation in video feeds
- Multi-person tracking in crowded scenes
- 3D human pose reconstruction from multiple cameras
- Gesture recognition and hand tracking applications
- Facial landmark detection for emotion analysis
- Motion capture and animation from video
Not Ideal For
- Projects deploying on low-power edge devices without GPU acceleration
- Applications requiring only single-person pose estimation with minimal computational overhead
- Teams needing a quick, out-of-the-box solution without complex dependency management
- Commercial products where licensing costs from FlintBox are prohibitive
Pros & Cons
Pros
Body keypoint estimation runtime is invariant to the number of people, enabling efficient processing in crowded scenes, as demonstrated in the runtime analysis comparison.
Detects 135 keypoints across body, face, hands, and feet, providing detailed pose data for advanced applications like motion analysis and gesture recognition.
Supports images, video, webcam, IP cameras, and custom sources, with outputs in JSON, XML, YML, or visual formats, allowing easy integration into various pipelines.
Runs on Ubuntu, Windows, macOS, and Nvidia TX2, with versions for CUDA, OpenCL, and CPU-only, ensuring compatibility across different environments.
Cons
Building from source requires managing dependencies like CUDA and OpenCV, which can be error-prone and time-consuming, as noted in the installation documentation.
Face and hand keypoint estimation runtime depends on the number of people, unlike body detection, which can degrade performance in scenes with many individuals, as admitted in the features section.
Non-commercial use is free, but commercial applications require purchasing a license from FlintBox, adding cost and administrative overhead for business deployments.
Frequently Asked Questions
Related Projects
face_recognitionThe world's simplest facial recognition api for Python and the command line
timmThe largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNetV4, MobileNet-V3 & V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more
detectron2Detectron2 is a platform for object detection, segmentation and other visual recognition tasks.
MLXMLX: An array framework for Apple silicon
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.