A Video Game Testing Method Utilizing Deep Learning

Computer video games must pass different types of tests before release. Yet most products in this multibillion-dollar industry still exhibit various compatibility problems when run on end consumers' computers. In this work, we propose a new automated testing method which utilizes deep convolutional neural networks to test video game compatibility with target runtime environments. This will result in better support for various computing environments that run video games and a reduction of the effort needed for testing them. Our method executes tests both on local computers and the cloud. Locally, a game tester will test the video game with normal testing routines. After that, these tests are automatically replicated on the cloud, running the video game on different environments. With the help of two convolutional neural networks, corrupted frames of the game containing artifacts are automatically discerned, and by comparing the local execution to the ones on the cloud, the corresponding problematic Draw Calls are determined. These are then used as a basis for comparison in order to determine the root cause of the graphical issue.