VR.net
收藏资源简介:
VR.net是由奥克兰大学、新加坡国立大学和Meta合作创建的一个大型、多样化且真实的虚拟现实游戏数据集,旨在推动基于机器学习的VR运动病研究。该数据集包含来自10种不同类型的10款真实世界VR游戏的约12小时游戏视频,每款游戏由至少5个不同用户评估。VR.net为每个视频帧提供了丰富的运动病相关标签,如相机/物体移动、景深和运动流等,这些标签通过自动工具从3D引擎的渲染管道中精确提取,无需访问游戏源代码。数据集的应用领域包括风险因素检测和舒适度预测,旨在改善VR体验并减少运动病的发生。
VR.net is a large, diverse and realistic virtual reality gaming dataset co-created by the University of Auckland, National University of Singapore and Meta, which aims to advance machine learning-based research on VR motion sickness. This dataset contains approximately 12 hours of gameplay video from 10 real-world VR games across 10 distinct categories, with each game evaluated by at least 5 different users. VR.net provides rich motion sickness-related labels for each video frame, such as camera/object movement, depth of field, motion flow and other relevant metrics. These labels are accurately extracted from the rendering pipeline of the 3D engine via automated tools, without requiring access to the game's source code. The application areas of this dataset include risk factor detection and comfort prediction, with the goal of improving VR experiences and reducing the incidence of motion sickness.



