基于Unreal Engine的自定义视频世界模型数据集
收藏资源简介:
该数据集是基于Unreal Engine渲染构建的定制化视频世界模型专用数据集,专门设计用于解决视频生成模型在长期场景一致性和记忆机制方面的挑战。数据集包含大量具有复杂场景遮挡和动态对象的长视频序列,并配备了精确的相机姿态标注,数据来源通过游戏引擎渲染生成确保了标注的准确性。创建过程通过定制化场景设计,系统性地引入了遮挡物和运动物体以模拟真实世界复杂环境。该数据集主要应用于视频世界模型的训练与评估,旨在解决现有方法在遮挡场景和动态对象处理中泛化能力不足的问题,为学习型上下文查询机制提供关键训练支撑。
This is a customized dataset dedicated to video world models, built using Unreal Engine rendering. It is specifically designed to address the challenges faced by video generation models in long-term scene consistency and memory mechanisms. The dataset contains a large number of long video sequences with complex scene occlusions and dynamic objects, and is equipped with precise camera pose annotations. As all data is generated via game engine rendering, the accuracy of the annotations is fully ensured. In the dataset construction process, customized scene designs were used to systematically introduce occluders and moving objects, simulating the complex real-world environments. This dataset is mainly used for the training and evaluation of video world models. Its core goal is to solve the problem of insufficient generalization ability of existing methods when dealing with occluded scenes and dynamic objects, providing critical training support for learned context query mechanisms.



