Project Description

We have developed a neural network to recognize the exact damaged area of the car from the uploaded image. We have labeled 50 000 car images that made it possible to detect: flat tires, broken windows, dent, vandalism, and crashes. Also, our machine learning model can recognize a brand of the car and determine from which angle the picture was taken. Our client used the model to create a mobile app, enabling users to insure their cars without the involvement of an insurance agent. 

Technologies:

• PyTorch based convolutional neural networks (Mask-R CNN) + scikit-learn / XGBoost for car properties and damage detection and price estimation.
• Scrapy for collecting data.