Cross-Perspective Annotated Dataset for Dynamic Object-Level Interest Modeling in Cloud Gaming
收藏资源简介:
本数据集是由南京邮电大学的研究人员创建的,旨在支持云游戏中的动态对象级兴趣建模。数据集包含来自《侠盗猎车手V》的501个3分钟视频剪辑和1503个游戏图像。每个图像对应两个标注的JSON文件,一个包含手动标注的感兴趣对象信息,另一个包含图像中所有对象的信息。数据集根据玩家的游戏速度分为三个类别:静止、低速和高速。每个类别中,场景进一步分为城市、乡村和高速公路。数据集包含了22个对象类别,并分析了影响玩家兴趣的主要因素,包括玩家的游戏速度、对象的大小和对象的速度。
This dataset was developed by researchers from Nanjing University of Posts and Telecommunications to support dynamic object-level interest modeling in cloud gaming. It contains 501 3-minute video clips and 1503 game screenshots sourced from *Grand Theft Auto V*. Each image is paired with two annotated JSON files: one holds manually annotated information of objects of interest, while the other records details of all objects appearing in the image. The dataset is divided into three categories based on players' in-game speed: stationary, low-speed, and high-speed. For each category, scenarios are further categorized into three types: urban, rural, and highway. The dataset covers 22 object categories, and analyzes the core factors influencing players' interest, including players' in-game speed, object size, and object movement speed.




