YOLO Car Damage
Pet Project
Computer Vision with YOLO
Jul 2025
The Problem
Manual car damage assessment is time-consuming, subjective, and requires expert knowledge. Insurance and automotive industries need automated solutions.
The Solution
Two-stage detection system: YOLOv8s for damage detection (3 classes: dirt, scratch, dent) + ONNX-based severity classifier (low, medium, high). Includes CLI and API for integration.
Key Innovation:
- •3-class damage detection (dirt, scratch, dent)
- •Severity classification (low, medium, high)
- •ONNX models for fast inference
- •CLI and API interfaces
- •Real-time processing (~15ms)
- •GPU support (CUDA optional)
Results
Inference speed: ~15ms per image, Detection accuracy: 85%+ on test set, mAP@0.5: 0.469
└─ Detection examples: https://github.com/ilyasidk/car_damage_detector_yolo/tree/master/docs/examples
└─ Training results: Precision 0.454, Recall 0.596, mAP@0.5: 0.469
└─ Model: YOLOv8s trained on 3-class dataset
Tech Stack
Backend
- ├─Python
AI & Integrations
- ├─YOLOv8s
- ├─ONNX Runtime
- ├─PyTorch
- ├─OpenCV
Infrastructure
- ├─Image Processing
- ├─Model Optimization