An Anchor Free Car Damage Detection Method
Haoran Jin, Xinkuang Wang, Zhongcheng Wu · 2023
Automatic car damage assessment is an intriguing problem in the practice of artificial intelligence. With the help of car damage assessment algorithms, automobile insurance companies, car rental, and car-sharing businesses could attain automatic auxiliary loss assessment or identify the insurance fraud problem. It would save amounts of time and money to replace the manual examination process in traditional car damage assessment with computer-aided damage examination. In this paper, we introduce an anchor-free object detection method for auxiliary car damage assessment adopting a car damage dataset. We use the coordinate attention mechanism and focal loss design to get higher accuracy with fewer parameters and GFLOPs compared to the baseline model. On the test set, our model gets 59.2% AP50 and 39.9% AP, outperforming the baseline model by 5.5%, and 8.7%, respectively. And the method reduces parameters by about 1.42M and GFLOPs by about 1.18.