CSE498, Collaborative Design, Fall 2022
Computer Science and Engineering
Michigan State University

CSAA Insurance Group, headquartered in Walnut Creek, California, is one of the largest AAA insurers in the country. They offer home, auto, and other lines of insurance to seventeen million people in twenty-three states and the District of Columbia.

In the span of a few short years, the insurance industry has become information driven, meaning CSAA requires large quantities of data to compete. However, data is difficult to access due to how it is typically captured, privacy concerns, and the relative infrequency of losses.

Our Synthetic Image Generation via Random Noise system provides an easy-to-use web application for CSAA data scientists to generate realistic image datasets. Our robust system is able to generate an image dataset for any subject of interest while eliminating privacy concerns, as the generated images are completely synthetic and do not depict any real people or vehicles.

Our application is a one-stop shop for all dataset generation needs. Users submit a text description of the images to generate, select the number of images desired, and quickly receive the dataset.

In addition to generating useful synthetic datasets, our system enables CSAA data scientists to update current dataset generators, as well as create new synthetic image generators quickly and effortlessly.

CSAA data scientists use our datasets for a variety of purposes, including providing concern-free data to CSAA’s third-party partners, improving the claims process for losses without documentation, and general insurance data needs.

Our web application utilizes Python Flask along with HTML, CSS, and JavaScript. The Stable Diffusion and Glide Diffusion models that our software uses are trained and run in Amazon Web Services EC2 instances with dedicated GPUs. The resulting datasets are stored in an AWS S3 Bucket.