VR.Net: A real-world dataset for virtual reality motion sickness research
IEEE Transactions on Visualization and Computer Graphics (Proc. IEEE VR 2024)
Machine-learning work on VR motion sickness needs large, accurately labelled, real-world data, which had been missing. VR.Net fills that gap with 165 hours of gameplay from 100 commercial games across ten genres, evaluated by 500 participants, and assigns 24 motion-sickness-related labels — such as camera and object movement, depth of field, and motion flow — to every video frame. Rather than label by hand, the team built a tool that extracts this ground truth directly from 3D engines' rendering pipelines without needing game source code, and they demonstrate the dataset on tasks like risk-factor detection and sickness prediction. The work won Best Paper at IEEE VR 2024.
VR.Net is a large-scale, real-world dataset for studying virtual-reality motion sickness. Collected from commercial VR games across many participants, it pairs gameplay with labelled sickness signals, enabling machine-learning models that predict and help mitigate cybersickness. The work received the Best Paper Award at IEEE VR 2024.
BibTeX
@article{wen2024,
author = {E. Wen and C. Gupta and P. Sasikumar and M. Billinghurst and J. Wilmott and E. Skow and A. Dey and S. Nanayakkara},
title = {VR.Net: A real-world dataset for virtual reality motion sickness research},
journal = {IEEE Transactions on Visualization and Computer Graphics (Proc. IEEE VR 2024)},
year = {2024},
doi = {10.1109/TVCG.2024.3372044}
}