FURI | Summer 2023

Video Game-Based Domain Generalization for Computer Vision

This research project aims to provide a new challenge to Domain Generalization (DG) in computer vision by creating a novel benchmark dataset made from video game scenes. Research project benchmark uses transitional generations of video game graphics as image domains unlike existing benchmarks. The project involves data collection, data cleaning, and manual annotation. Once the dataset is constructed, DG baselines of our dataset with existing DG methods will be provided. Dataset and baselines will be shared to the public for the research community to invite further study in DG on video game graphics.