Skip to content

Latest commit

 

History

History
102 lines (72 loc) · 3.81 KB

README.md

File metadata and controls

102 lines (72 loc) · 3.81 KB
Raw

🤖 Project Summary: AI Rubik's Cube Recognition Using TensorFlow.js and RetinaNet-SpineNet-49

This project utilizes TensorFlow.js and the RetinaNet-SpineNet-49 model (retinanet_spinenet_mobile_coco) to train an AI system that recognizes and interprets Rubik's Cube patterns via a camera. By integrating computer vision 🖼️, COCO annotations 🗂️, and state-of-the-art object detection models 🚀, the system detects and identifies each face and tile color of a Rubik's Cube, paving the way for automated solving.

📝 Description

The project is structured into three core components:

  1. Data Preparation 🖌️:

    • Collected Rubik's Cube images annotated using LabelMe.
    • Converted annotations from LabelMe JSON format to COCO JSON format using a custom Python script.
    • Defined detection categories, including color tiles (red_tile, white_tile, blue_tile, etc.).

  2. Model Training 🏋️‍♀️:

    • This project uses the RetinaNet-SpineNet-49 architecture, specifically the retinanet_spinenet_mobile_coco model provided by TensorFlow as the base model.
    • Fine-tuned the model for real-time classification and localization of Rubik's Cube tiles.

  3. Visualization and Evaluation 👀:

    • Developed a visualization script to overlay predictions on test images with bounding boxes and labels.
    • Exported results to PNG format for validation and analysis.

✨ Key Features

  • Advanced Object Detection 🎯: High-precision Rubik's Cube tile detection using the RetinaNet-SpineNet-49 model.
  • Dynamic Tile Recognition 🕹️: Real-time identification of cube tiles and colors via camera input.
  • Streamlined Annotation Workflow 🔄: Seamless conversion of LabelMe JSON annotations into the COCO JSON format.
  • Custom Visualization Tools 🖼️: Debugging and enhancing predictions through overlayed visual outputs.

⚙️ How It Works

  1. Data Collection: Annotate Rubik's Cube images using LabelMe. ✍️
  2. COCO Conversion: Convert annotations with the labelme_to_coco.py script. 🔄
  3. Model Training:
    • Configure the retinanet_spinenet_mobile_coco model using TensorFlow Model Garden.
    • Train the model on a custom Rubik's Cube dataset. 🧠
  4. Deployment: Use TensorFlow.js to deploy the trained model for real-time detection. 🌐

🚀 Getting Started

Prerequisites 🛠️

  • Python 3.8+
  • TensorFlow
  • TensorFlow.js
  • TensorFlow Model Garden
  • LabelMe for annotation

Installation 📦

  1. Clone the repository:

    git clone [repository_link]
cd rubiks-cube-detection

  2. Install dependencies:

    pip install -r requirements.txt

  3. Annotate your Rubik's Cube images using LabelMe and place them in the images/ folder.

Usage 💻

  1. Convert Annotations:

    python labelme_to_coco.py

  2. Train the Model:

    • Modify the configuration for retinanet_spinenet_mobile_coco.
    • Train the model: 
      python train.py --model=retinanet_spinenet_mobile_coco --config=configs/retinanet_spinenet_mobile_coco.config --data_dir=path_to_coco_data

  3. Run Detection:

    • Convert the trained model to TensorFlow.js format.
    • Deploy the model for real-time detection.

📊 Results

Sample visualizations with bounding boxes and labels are saved in the outputs/ folder. 🖼️

🛤️ Roadmap

  • 🤖 Integrate a Rubik's Cube-solving algorithm.
  • 📈 Expand the dataset to improve detection accuracy.
  • 🌐 Deploy the system as a web-based application.

🤝 Contributions

Contributions are welcome! Feel free to submit issues or pull requests for improvements. 💡