From Synthetic Data to Real-World Efficiency: Automated Vehicle Damage Detection and Vehicle Parts Segmentation.

This project focuses on developing an automated system for vehicle damage detection and vehicle parts segmentation, utilizing synthetic data to improve real-world efficiency. The dataset, SYNDCAR, consists of 245 high-resolution images with annotations for both vehicle damage (broken glass, broken lights, cracks, scratches) and vehicle parts (28 different components). The project aims to automate the identification of damages and the segmentation of vehicle parts to streamline vehicle inspections. Three different models are employed: YOLOv9, YOLO26, and DA-ViT-YOLO (Domain-Adaptive Vision Transformer YOLO). The models are trained on the SYNDCAR dataset and evaluated for their performance on detecting damages and segmenting vehicle parts. The use of synthetic data, combined with these advanced models, allows the system to overcome the limitations of traditional vehicle inspection systems. By focusing on domain adaptation and leveraging state-of-the-art detection algorithms, this system provides an efficient, scalable, and automated solution for vehicle damage detection and parts segmentation, paving the way for enhanced vehicle maintenance processes and cost-effective solutions for the automotive industry.

Keywords: Automated Vehicle Damage Detection, Vehicle Parts Segmentation, YOLOv9, YOLO26, DA-ViT-YOLO, Synthetic Data, Computer Vision, Domain Adaptation.