Visual Regression Testing of Video Games Using Foundation Models

During visual regression testing of video games, automated techniques are employed to identify visual bugs. Visual bugs can be detected by just looking at them, such as those related to texture and lighting, but also more complex ones like those related to the physics engine (such as a flying horse) that require common-sense reasoning. The impact of visual bugs varies from minor irritations to rendering a game unplayable. Although human experts can often easily spot these bugs, automated detection is challenging. As a result, visual regression testing usually requires significant resources for manual testing.

This project proposes leveraging foundation models to address the visual regression testing problem. Foundation models are large-scale machine learning models that are pre-trained on very large amounts of data from different domains. This project investigates how foundation models can be leveraged and fine-tuned for visual regression testing of games that were published by Electronic Arts (EA), a world-leading game publisher with over 8,000 game makers. Automated techniques for detecting visual bugs in video games would help reduce the cost of testing for EA. The project’s goal is not to replace testers but to help them find more bugs in less time.

Faculty Supervisor:

Cor-Paul Bezemer

Student:

Partner:

Electronic Arts Canada (Burnaby, BC)

Discipline:

Computer science

Sector:

Information and cultural industries

University:

University of Alberta

Program: