🚗 Advanced Car Damage Detector AI

An AI-powered system for automatic car damage detection using SAM + CNN + SSIM hybrid technology.

Example: Before and after comparison with AI-detected damage highlighting

✨ Key Features

• 🔍 Multi-Stage AI Detection: Combines SAM segmentation, CNN classification, and SSIM analysis
• 📸 Before/After Comparison: Automatic damage detection from vehicle photos
• ⚡ Fast Processing: Results in 3-5 seconds per image pair
• 📊 Confidence Scoring: Detailed damage assessment with probability scores
• 🎨 Smart Visualization: Automatic damage highlighting and annotation
• 📱 Web Interface: User-friendly Streamlit application

🏗️ Architecture

The system uses a multi-stage approach:

1. Image Alignment: SIFT keypoint matching for perspective correction
2. Change Detection: SSIM structural similarity analysis
3. Segmentation: SAM (Segment Anything Model) for region identification
4. Classification: CNN filtering to distinguish damage from non-damage
5. Assessment: Overall severity and confidence scoring

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• CUDA-capable GPU (optional, for faster processing)

Installation

1. Clone the repository

   git clone <repository-url>
   cd car-damage-detector

2. Create virtual environment

   python -m venv venv
   
   # On Windows
   venv\Scripts\activate
   
   # On macOS/Linux
   source venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Download SAM model (optional)

   # Download SAM checkpoint for enhanced segmentation
   wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth
   # Place in project root or specify path in configuration

Running the Application

1. Start the web interface

   streamlit run app.py

2. Open your browser and navigate to http://localhost:8501

3. Upload photos:

   • Before photos: Car in good condition
   • After photos: Car after use/rental

4. Click "Start Advanced Damage Detection" to analyze

📖 Usage Guide

Basic Workflow

1. Navigate to "Damage Detection" page
2. Upload before photos (car in good condition)
3. Upload after photos (car after use/rental)
4. Start analysis with the detection button
5. Review results with damage highlighting and confidence scores

Best Practices

• 📸 Photo Quality: Use high-resolution images (1920x1080 or higher)
• ☀️ Lighting: Ensure consistent lighting conditions
• 🚗 Framing: Include the entire vehicle in the frame
• 📐 Angles: Take photos from similar angles when possible
• 🕐 Timing: Capture photos at similar times of day

Configuration Options

The system allows fine-tuning of detection parameters:

# Adjust detection sensitivity
detector.set_damage_region_padding(0.4, 100)  # CNN context padding
detector.set_visualization_padding(0.5, 60)   # Visual marking padding

# Modify thresholds
detector.ssim_threshold = 0.75        # Change detection sensitivity
detector.confidence_threshold = 0.579 # CNN confidence threshold

🧠 Model Training

Training Your Own CNN Classifier

1. Prepare training data in the training_data/ directory:

   training_data/
   ├── damage/
   │   ├── image1.jpg
   │   ├── image2.jpg
   │   └── ...
   └── no_damage/
       ├── image1.jpg
       ├── image2.jpg
       └── ...
   

2. Run training script

   python train_damage_classifier.py

3. Monitor training with real-time metrics and early stopping

4. Model saved to models/best_damage_classifier.pth

Training Parameters

• Architecture: ResNet18 with custom classification head
• Input Size: 256x256 pixels
• Augmentation: Random flips, brightness, contrast adjustments
• Optimization: Adam optimizer with learning rate scheduling
• Early Stopping: Training stops at 90% accuracy or max epochs

🛠️ Technical Details

Core Dependencies

• PyTorch 2.0+: Deep learning framework
• OpenCV: Computer vision operations
• SAM: Segment Anything Model for segmentation
• scikit-image: Image processing algorithms
• Streamlit: Web interface framework

File Structure

car-damage-detector/
├── app.py                      # Main Streamlit application
├── damage_detector.py          # Core AI detection engine
├── damage_visualizer.py        # Damage visualization utilities
├── train_damage_classifier.py  # CNN training script
├── damage_dataset.py           # Dataset handling
├── requirements.txt            # Python dependencies
├── models/                     # Trained models directory
├── training_data/              # Training dataset
└── README.md                   # This file