Omnidirectional Blender 3D (OB3D)
收藏资源简介:
OB3D数据集由东北大学和国立清华大学联合创建,是一个用于评估从多方位全向图像进行3D重建方法的新合成数据集。数据集包含12个室内外场景,每个场景提供了RGB图像、深度图、法线图、稀疏3D点云和真实相机参数,使得可以定量评估3D重建方法的性能。OB3D数据集适用于相关任务,如新视角合成和相机校准,有助于从全向图像开发3D重建技术。数据集旨在解决现有数据集缺乏评估全向图像3D重建的特定环境问题,通过提供精确的相机参数和丰富的场景信息,OB3D为3D重建方法提供了标准化和公平的评估基准。
The OB3D dataset, jointly developed by Northeast University and National Tsinghua University, is a novel synthetic dataset for evaluating 3D reconstruction methods using omnidirectional images from multiple viewpoints. The dataset contains 12 indoor and outdoor scenes, with each scene providing RGB images, depth maps, normal maps, sparse 3D point clouds, and ground-truth camera parameters, enabling the quantitative evaluation of 3D reconstruction method performance. The OB3D dataset is applicable to related tasks such as novel view synthesis and camera calibration, and facilitates the development of 3D reconstruction techniques based on omnidirectional images. Designed to address the critical limitation of existing datasets that lack dedicated environments for evaluating 3D reconstruction from omnidirectional images, OB3D offers a standardized and fair evaluation benchmark for 3D reconstruction methods by providing accurate camera parameters and rich scene information.




