AmirEzze/diamond-color-grading

💎 Complete Diamond Grading System

An AI-powered system that analyzes diamond images and predicts Shape, Color Grade, and Clarity using deep learning.

🎯 Model Performance

📊 What We Classify

🔷 Shape (8 Types)

• —Round, Oval, Cushion, Princess
• —Emerald, Pear, Heart, Marquise

🎨 Color (4 Grades)

• —D-F: Colorless (premium)
• —G-J: Near Colorless (excellent value)
• —K-M: Faint (good value)
• —N-R: Very Light (budget)

✨ Clarity (3 Levels)

• —High: FL, IF, VVS1, VVS2 (eye-clean)
• —Medium: VS1, VS2, SI1 (excellent value)
• —Low: SI2, I1, I2, I3 (visible inclusions)

🚀 Usage

Upload a clear, well-lit image of a diamond (preferably top-down view) and get instant predictions for all three attributes!

🔧 Technical Details

• —Architecture: EfficientNetB2 (pretrained on ImageNet)
• —Framework: TensorFlow/Keras 3.6.0
• —Input: 224×224 RGB images
• —Training Data: 48,726 diamond images
• —Inference Time: ~300ms for all 3 predictions

📈 Model Files

• —shape_model.keras (35 MB) - Shape classifier
• —color_model.keras (30 MB) - Color classifier
• —clarity_model.keras (35 MB) - Clarity classifier
• —JSON files with class metadata

⚠️ Limitations

• —Best for standard cuts with good lighting
• —Not a replacement for professional GIA/AGS grading
• —Works best with top-down diamond views
• —Color grading most accurate on white backgrounds

📄 License

MIT License - Free for commercial and personal use

Made with ❤️ for the gemology community