Good-GYM: AI Fitness Assistant 💪

AI Fitness Assistant

English | 中文

🆕 Changelog

• 2025-06-07: Major update! Dropped YOLO models and all GPU support. Now uses only RTMPose for pose detection, and runs on CPU only. Simpler, more compatible, and easier to use.
• 2025-06-12：Optimize exercise_counters.py for counting accuracy, code structure optimization
• 2025-11-14: Reverted asynchronous pose detection due to accuracy issues, restored synchronous pose detection. Fixed crash when switching from statistics mode back to detection mode.
• 2025-11-15: New exercise database feature! All exercise configurations are now managed in data/exercises.json file. You can easily add, modify, or remove exercise types without modifying code.
• 2026-03-04: Add optional GPU acceleration support, now supports NVIDIA GPUs
• 2026-03-28: Add mobile application support, now supports IOS

🔮 Future Development

• Multi-language interface
• Improve pose detection accuracy
• Add support for more exercise types
• Add custom exercise types template
• Recognizing Motion Accuracy
• Mobile Application Support
• Motion Error Correction Indication
• Add voice feedback

---

📱 Try Good-GYM on iOS

---

🌟 Features

• Real-time Exercise Counting - Automatically counts your repetitions
• Multiple Exercise Support - Including squats, push-ups, sit-ups, bicep curls, and many more
• Advanced Pose Detection - Powered by RTMPose for accurate tracking
• CPU & GPU Support - Works on CPU by default, with optional GPU acceleration
• Visual Feedback - Live skeleton visualization with angle measurements
• Workout Statistics - Track your progress over time
• User-friendly Interface - Clean PyQt5 GUI with intuitive controls
• Works with any webcam - No special hardware required
• Runs locally - Complete privacy

📦 Direct Download

📱 iOS App

💻 Windows Desktop

• If you don't want to set up a Python environment, you can download our pre-packaged executable:

  Baidu Netdisk Link code: 8866

  Google Drive

📝 Usage Guide

🎯 Custom Exercise Types

All exercise types are now stored in the data/exercises.json file. You can easily add, modify, or remove exercise types without modifying code!

How to Add a New Exercise Type

1. Keypoint Index Reference

   • The system uses COCO 17 keypoint format:

                    ○ 0
                   /|\
            1 ●     |     ● 2
           3 ●      |      ● 4
                    |
          5 ●———————————● 6
            |       |       |
            |       |       |
          7 ●       |       ● 8
            |       |       |
            |       |       |
          9 ●       |       ● 10
                    |
         11 ●———————————● 12
            |               |
            |               |
         13 ●               ● 14
            |               |
            |               |
         15 ●               ● 16
     
      Index │ Keypoint    Index │ Keypoint
      ──────┼──────────  ──────┼──────────
        0   │ Nose          9  │ L.Wrist
        1   │ L.Eye        10  │ R.Wrist
        2   │ R.Eye        11  │ L.Hip
        3   │ L.Ear        12  │ R.Hip
        4   │ R.Ear        13  │ L.Knee
        5   │ L.Shoulder   14  │ R.Knee
        6   │ R.Shoulder   15  │ L.Ankle
        7   │ L.Elbow      16  │ R.Ankle
        8   │ R.Elbow
     

2. Configuration Parameters

   • down_angle: Angle threshold when lowering (degrees)
   • up_angle: Angle threshold when raising (degrees)
   • keypoints.left: Left side three keypoint indices [pt1, pt2, pt3] for angle calculation
   • keypoints.right: Right side three keypoint indices [pt1, pt2, pt3] for angle calculation
   • is_leg_exercise: Whether it's a leg exercise (true/false), affects counting logic
   • angle_point: Keypoint indices [pt1, pt2, pt3] for displaying angle lines on video

3. Example: Adding a New Exercise

   "my_custom_exercise": {
     "name_zh": "我的自定义运动",
     "name_en": "My Custom Exercise",
     "down_angle": 120,
     "up_angle": 170,
     "keypoints": {
       "left": [5, 7, 9],
       "right": [6, 8, 10]
     },
     "is_leg_exercise": false,
     "angle_point": [6, 8, 10]
   }

4. Restart the Application

📋 Requirements

• Python 3.9
• Webcam
• Windows/Mac/Linux: Works on CPU by default. Optional GPU acceleration available for source deployments.

🚀 Environment Setup

Installation

1. Clone and install

   git clone https://github.com/yo-WASSUP/Good-GYM.git
   cd Good-GYM
   
   # Create virtual environment
   python -m venv venv
   # Activate (Windows)
   .\venv\Scripts\activate
   # or (Mac/Linux)
   source venv/bin/activate
   
   # Install dependencies
   pip install -r requirements.txt

2. Run the application

   python run.py

GPU Acceleration (Optional)

If you have an NVIDIA GPU, you can enable GPU-accelerated inference for better performance.

Prerequisites: NVIDIA GPU + NVIDIA Driver installed

Resource Usage:

• CUDA runtime libraries require ~3 GB disk space
• Models use only ~200 MB VRAM — any NVIDIA GPU with 2GB+ VRAM will work

# 1. Replace onnxruntime with the GPU version
pip uninstall onnxruntime
pip install onnxruntime-gpu

# 2. Install CUDA runtime libraries via pip (no need to install CUDA Toolkit manually)
pip install nvidia-cudnn-cu12 nvidia-cublas-cu12 nvidia-cuda-runtime-cu12 nvidia-cufft-cu12 nvidia-curand-cu12 nvidia-cusolver-cu12 nvidia-cusparse-cu12 nvidia-cuda-nvrtc-cu12

The application will auto-detect GPU availability at startup. You can toggle GPU on/off via the "GPU Acceleration" switch in the control panel.

Note: The pre-packaged EXE only supports CPU mode. GPU acceleration is only available when running from source.

🖼️ Screenshots

Desktop

iOS App

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Thanks to RTMPose open source pose detection model: https://github.com/Tau-J/rtmlib

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

⭐ Star History