Dexterous World Models (DWM) dataset
收藏资源简介:
Dexterous World Models (DWM)数据集由首尔国立大学的研究团队构建,旨在支持静态3D场景中灵巧手部动作驱动的视觉动态建模。该数据集包含合成的自我中心视角交互视频、静态场景视频和手部视频,以及固定摄像头的真实世界交互视频,以提供精确对齐和多样化的交互动态。数据集的构建过程涉及从静态3D场景渲染和手部运动序列生成对齐的交互视频,以训练视频扩散模型。该数据集主要应用于交互式数字孪生和视觉世界建模领域,旨在解决静态3D场景中灵巧手部动作驱动的动态变化预测问题。
The Dexterous World Models (DWM) dataset was constructed by a research team from Seoul National University, aiming to support visually dynamic modeling driven by dexterous hand movements in static 3D scenes. This dataset includes synthetic egocentric interactive videos, static scene videos, hand videos, as well as real-world interactive videos captured by fixed cameras, to provide precisely aligned and diverse interaction dynamics. The construction process of the dataset involves rendering static 3D scenes and generating aligned interactive videos via hand motion sequences, for training video diffusion models. This dataset is mainly applied in the fields of interactive digital twins and visual world modeling, aiming to address the problem of predicting dynamic changes driven by dexterous hand movements in static 3D scenes.

- 1Dexterous World Models首尔国立大学 · 2025年



