Personal Yoga Coach (On-Device AI)

Personal Yoga Coach is a cutting-edge, AI-powered application designed to provide personalized yoga guidance directly on your device. The app enhances your yoga practice by offering real-time feedback, customized routines, and on-demand coaching—all while operating entirely offline to ensure privacy and accessibility.

This project won the Meta Consumer AI on the Edge Hackathon in December 2024. Huge thanks to the Lightning-Fast Teammates for making this possible: Gaspard Hassenforder, Alex Bohane, Lazlo Guioland, Julien Roubaud, and Alba Maria Tellez Fernandez.

---

🌟 Features

• 🖥️ On-Device AI: Runs offline using Qwen/Qwen2.5-0.5B-Instruct for local LLM inference, ensuring privacy and no dependency on the internet.
• 🧘 Real-Time Feedback: AI-powered pose correction and accuracy scoring to help improve posture and technique.
• 🗣️ Voice and Visual Guidance: Provides both verbal cues and on-screen visual overlays for pose corrections.
• 🦾 Computer Vision-Powered Analysis: Uses MediaPipe and OpenCV for real-time human pose detection and skeleton tracking.
• 🎙️ AI-Driven Voice Feedback: NLP-driven corrections via Qwen LLM, using structured prompt engineering for precise feedback. Integrated with Piper TTS for voice-based corrections.
• 📊 Adaptive Feedback Mechanism: Scores pose accuracy and dynamically adjusts the difficulty level and feedback precision based on user performance.
• 🖥️ Gradio UI for User Interaction: Browser-based interactive interface that provides real-time visualization of detected poses and feedback.
• ⚡ Optimized Performance: Implements multi-threaded pose analysis, caching mechanisms, and lightweight inference models for seamless execution on local devices.

---

🔧 How It Works

1. Pose Detection:

   • Utilizes MediaPipe's BlazePose for extracting human body key points.
   • Processes the skeletal structure using OpenCV overlays.

2. AI-Powered Correction:

   • Compares detected poses with predefined yoga posture templates.
   • Uses Qwen LLM to provide textual and audio-based corrections.

3. Real-Time Feedback:

   • Processes corrections dynamically and offers voice guidance via Piper TTS.
   • Displays visual cues on Gradio UI for better understanding.

4. Adaptive Accuracy Scoring:

   • Computes pose similarity scores using cosine similarity between detected and ideal skeletal points.
   • Adjusts guidance based on user performance over multiple sessions.

---

🛠️ Installation Guide

Follow these steps to install and run Personal Yoga Coach on your local machine.

Prerequisites

• Python 3.8 or higher
• pip
• Virtual environment support (optional but recommended)

Installation Steps

1. Clone the Repository:

   git clone https://github.com/LazGL/YogaLivePoseCorrector.git

2. Navigate to the Project Directory:

   cd YogaLivePoseCorrector

3. Create a Virtual Environment (Recommended):

   python -m venv venv

4. Activate the Virtual Environment:

   • On Windows:

     .\venv\Scripts\activate

   • On macOS/Linux:

     source venv/bin/activate

5. Install Dependencies:

   pip install -r requirements.txt

6. Run the Application:

   python app_llm.py

---

🚀 Usage

1. Launch the application following the installation steps.
2. Open the Gradio UI in your browser.
3. Follow the on-screen instructions to start your yoga session.
4. Receive real-time feedback through voice and visual guidance.
5. Adjust your posture based on the corrections provided by the AI.

---

🛡️ Privacy & Security

• All processing occurs entirely on-device, ensuring that your data never leaves your system.
• No internet connection is required, providing a secure and private experience.
• Your movement data and corrections remain completely confidential.

---

🤝 Contributing

We welcome contributions to enhance Personal Yoga Coach! If you'd like to contribute:

1. Fork the repository.
2. Create a new branch (git checkout -b feature-branch).
3. Commit your changes (git commit -m "Add new feature").
4. Push to the branch (git push origin feature-branch).
5. Open a pull request.

---

📜 License

This project is licensed under Apache 2.0, allowing modification and redistribution, as long as proper credit is given to the original author.

🚨 Additional Restriction: This code may not be used for commercial purposes without permission, as detailed in NON-COMMERCIAL.md.

Any improvements based on this code must give proper credit to the original author.

---

📞 Support

For any issues or suggestions, feel free to open an issue on GitHub or contact us via email.

🔗 GitHub Repository: YogaLivePoseCorrector