| Property | Value |
|---|---|
| License | Apache 2.0 |
| Input Resolution | 256x256 |
| Detector Parameters | 815K (3.14 MB) |
| Landmark Detector Parameters | 3.37M (12.9 MB) |
| Paper | BlazePose: On-device Real-time Body Pose tracking |
What is MediaPipe-Pose-Estimation?
MediaPipe-Pose-Estimation is a sophisticated machine learning pipeline designed specifically for mobile deployment, capable of detecting and tracking human body poses in real-time from images and video streams. The model implements a two-stage architecture, consisting of a pose detector and a landmark detector, optimized for efficiency on Qualcomm devices.
Implementation Details
The model architecture consists of two main components: MediaPipePoseDetector (815K parameters) and MediaPipePoseLandmarkDetector (3.37M parameters). Both components are optimized for FP16 precision and primarily utilize the NPU (Neural Processing Unit) for computation. The implementation supports multiple runtime formats including TFLite and ONNX, with impressive inference times ranging from 0.5ms to 2ms on modern devices.
- Dual-stage detection pipeline for accurate pose estimation
- Optimized for mobile deployment with FP16 precision
- Supports both TFLite and ONNX runtime formats
- Efficient memory usage with peak consumption varying by device
Core Capabilities
- Real-time pose detection and tracking
- Human body landmark detection with high accuracy
- Efficient performance on mobile devices
- Cross-platform compatibility
- Memory-efficient implementation
Frequently Asked Questions
Q: What makes this model unique?
The model's dual-stage architecture and optimization for mobile devices make it particularly efficient for real-time applications. Its ability to run on NPUs with low inference times (sub-millisecond in many cases) while maintaining high accuracy sets it apart from other pose estimation models.
Q: What are the recommended use cases?
The model is ideal for mobile applications requiring real-time pose estimation, including fitness apps, motion tracking, augmented reality applications, and gesture-based interfaces. Its optimization for Qualcomm devices makes it particularly suitable for Android applications requiring efficient pose detection.