Letsapatiiso07/pothole-detector

AI-powered pothole detection system with synthetic data generation for African road conditions # 🚗 Pothole & Road Anomaly Detector A production-ready computer vision system for real-time pothole and road anomaly detection using deep learning. Specifically enhanced with synthetic data generation to handle challenging African road conditions including dust, rain, mud, and harsh sunlight. ## 🌟 Key Features ### 🎯 Core Capabilities - **Transfer Learning**: Leverage pre-trained ResNet50, EfficientNet, or MobileNetV2 - **Grad-CAM Visualization**: Interpretable AI showing exactly where potholes are detected - **Multi-Dataset Support**: Easy integration and merging of multiple pothole datasets ### 🌍 African Road Condition Simulation - **Dust Effects**: Brownish haze and airborne particles common on unpaved roads - **Rain Simulation**: Wet surfaces, water reflections, and rain streaks - **Mud Splatter**: Realistic mud spots and droplets on camera lens - **Harsh Sunlight**: Overexposure, lens flare, and extreme brightness - **Dynamic Shadows**: Irregular patterns from trees, buildings, and power lines - **Worn Surfaces**: Aged road texture with reduced saturation ### 📊 Comprehensive Analytics - **Accuracy, Precision, Recall, F1-Score**: Complete performance metrics - **AUC-ROC Curve**: Threshold-independent evaluation - **Confusion Matrix**: Visual performance breakdown - **Training Curves**: Real-time monitoring via TensorBoard - **Classification Reports**: Detailed per-class statistics ## 🚀 Quick Start ### Dataset Setup **Option 1: Download Kaggle Dataset** ```bash # Download from Kaggle kaggle datasets download -d atulyakumar98/pothole-detection-dataset # Extract unzip pothole-detection-dataset.zip -d data/ # Verify structure python dataset_checker.py --data_dir data/pothole-detection-dataset ``` **Option 2: Create Sample Dataset** ```bash # Generate synthetic test data python create_sample_data.py --output_dir data/pothole-detection-dataset --num_samples 100 ``` ### Training ```bash # Basic training python main.py --mode train \ --data_dir data/pothole- …