Automobile Insurance Processing using Deep Convolutional Networks

We focus on automating the task of automobile insurance processing using Deep Convolution Networks, due to the limited data we find that using transfer learning and using variational Auto encoders to find features works well, we have created four models, classification of car or not, classification of whether car is damaged or not, classification of where the damage has occurred and the severity of the damage respectively. We show different methods that can be used in performing car damage analysis and this paper is not an exhaustive search over the entire domain car damage insurance and claims processing. This paper records the different techniques we have employed in order to analyze the car damage.