Car Damage Detection System
Automated car damage classification using deep learning and computer vision
Domain / Function
Computer Vision & Deep Learning
Project Overview
Developed a deep learning–based Proof of Concept to automatically detect and classify car damage from images. The system categorizes damage into six classes covering front and rear sections of vehicles.
The solution leverages transfer learning using a pre-trained ResNet50 model and demonstrates how deep learning can significantly reduce manual inspection time in insurance and automobile assessment workflows.
Real-time Damage Prediction 1
System workflow
Model Evaluation
Key Features
- Six-class car damage classification (Front/Rear – Normal, Breakage, Crushed)
- Transfer learning using ResNet50 (ImageNet pre-trained)
- Image preprocessing and augmentation for robustness
- Achieved 80.52% validation accuracy (exceeding target)
- Confusion matrix–based model evaluation
- Interactive Streamlit interface for real-time predictions
Project Details
The dataset consisted of 2300 labeled images distributed across six damage categories. Multiple architectures were tested, including custom CNNs and EfficientNet, before selecting ResNet50 for its superior feature extraction and performance.
This project highlights strong skills in deep learning workflows, transfer learning, model evaluation, and real-world deployment through a user-friendly web interface.