Categorizing Exterior Damage in Car Using Deep Learning Techniques

Abstract

Recently, the assembly of picture-based automobile insurance is a crucial part with considerable purview for automation reach. In this research work, we deal with the difficulty of classifying car damage, where a number of divisions in categorizing the level could also be fine-granular. To the present reason, we are exploring techniques based on deep learning. Initially we attempt to train a CNN directly with a set of coaching data. However, it isn't working well thanks to a little collection of labeled data. Hence, we investigate the domain-specific pre-training effect amid fine-tuning with an outsized number of annotated training-data. As Faster R-CNN and SVM haven't identified damaged cars with high accuracy, and therefore the Cascade R-CNN takes an immense amount of your time to coach and check the info that we are performing on to suit. Hence, we are training data into an R-CNN Mask that produces adequate results compared to traditional Neural Networks. Though there's tons of unknown like partial images, hence the classifier was built to detect amorphous damages. The model is layered over 3 classifications of detecting the car and examining whether the damage dealt is high or low. Finally, the classifier is projected with the flask environment to form the working experience easier, since it runs on a local host the compile time doesn't exceed 5 seconds regardless of the standard of the image. Experimental results indicate that Mask R-CNN works better than convolutional R-CNN as transfer learning is way more better when compared with area specific fine-tuning like Cascade and Faster R-CNN. We achieve 89.5 per cent accuracy with Mask R-CNN through transfer combination.