YOWO Dataset
收藏资源简介:
YOWO数据集是由富士通研究所构建的首个面向室内场景建模与天花板相机标定的协同采集基准数据集。该数据集通过AI2THOR和Gym-UnrealCV仿真引擎生成,包含1个头部RGB-D视角与5-17个天花板相机视角的同步视频流,涵盖低中高三种共视度场景。数据采集过程模拟移动代理遍历室内环境,同步记录自我中心视角的深度信息与多视角RGB观测,创新性地融合静态场景特征与移动关键点。本数据集主要应用于室内三维重建、多相机标定、视觉定位等领域,旨在解决传统视觉定位方法在视角差异下的特征歧义问题。
The YOWO dataset, constructed by Fujitsu Research Institute, is the first joint acquisition benchmark dataset for indoor scene modeling and ceiling camera calibration. Generated using the AI2THOR and Gym-UnrealCV simulation engines, this dataset includes synchronized video streams from 1 egocentric RGB-D viewpoint and 5 to 17 ceiling camera viewpoints, covering three types of scenes with low, medium, and high co-visibility. The data collection process simulates a mobile agent navigating through indoor environments, synchronously recording egocentric depth information and multi-view RGB observations, and innovatively fuses static scene features and moving keypoints. This dataset is primarily applied in fields such as indoor 3D reconstruction, multi-camera calibration, visual localization, and other related domains, aiming to address the feature ambiguity issue of traditional visual localization methods under varying viewpoint differences.




