Faultiz - Visual Defect Classifier & Vision LLM Assistant

A comprehensive AI-powered solution for industrial defect detection, analysis, and repair recommendations.

🚀 Features

• CNN-based Defect Classification: ResNet/EfficientNet backbone trained on industrial defect datasets
• Grad-CAM Explainability: Visual heatmaps showing defect locations
• Google Cloud Vision OCR: Text extraction from defective items (serial numbers, labels)
• Hugging Face LLM Integration: AI-powered repair suggestions
• Streamlit UI: User-friendly interface for complete workflow

🏗️ Architecture

Image → CNN Classifier → Grad-CAM → Defect Label
   ↓
OCR API → Text Extraction
   ↓
Defect + OCR → LLM → Repair Suggestions
   ↓
Streamlit Dashboard

📁 Project Structure

faultiz/
├── app/
│   ├── __init__.py
│   ├── main.py              # Streamlit main app
│   └── components/          # UI components
├── models/
│   ├── __init__.py
│   ├── defect_classifier.py # CNN model
│   ├── explainer.py         # Grad-CAM implementation
│   └── pretrained/          # Model weights
├── services/
│   ├── __init__.py
│   ├── ocr_service.py       # Google Cloud Vision
│   ├── llm_service.py       # Hugging Face LLM
│   └── inference.py         # Main inference pipeline
├── utils/
│   ├── __init__.py
│   ├── image_processing.py
│   ├── config.py
│   └── data_loader.py
├── data/
│   ├── sample_images/
│   └── defect_classes.json
├── requirements.txt
├── config.yaml
└── setup.py

🛠️ Installation

1. Clone the repository
2. Install dependencies: pip install -r requirements.txt
3. Set up Google Cloud credentials
4. Run the app: streamlit run app/main.py

🔧 Configuration

• Update config.yaml with your API keys and model paths
• Place Google Cloud service account key in credentials/

📊 Supported Defect Types

• Scratches
• Dents
• Cracks
• Corrosion
• Missing Parts
• Color Defects
• Shape Anomalies

🤖 Models Used

• CNN Backbone: EfficientNet-B0 (customizable)
• LLM: Hugging Face Transformers (configurable model)
• OCR: Google Cloud Vision API

📝 License

MIT License