GlassRecon
收藏资源简介:
GlassRecon是由南方科技大学和香港科技大学联合创建的室内RGB-D数据集,专门针对玻璃表面深度估计问题。该数据集包含917张带有玻璃结构的室内场景图像,每张图像均提供RGB信息、原始深度图、玻璃掩码、几何校正后的真实深度及相机内参。数据来源于Matterport3D和RGB-D GSD公开数据集,通过多步骤几何标注流程生成,包括深度点标注、3D平面拟合和射线-平面交会计算。该数据集旨在解决机器人导航中玻璃表面深度感知不准的难题,为深度补全和场景重建提供标准化评测基准。
GlassRecon is an indoor RGB-D dataset jointly developed by Southern University of Science and Technology and The Hong Kong University of Science and Technology, specifically tailored for the depth estimation task on glass surfaces. This dataset contains 917 indoor scene images with glass structures, where each image is accompanied by RGB information, raw depth maps, glass masks, geometrically corrected ground-truth depth maps, and camera intrinsic parameters. Sourced from the public datasets Matterport3D and RGB-D GSD, the dataset is generated through a multi-step geometric annotation workflow, including depth point annotation, 3D plane fitting, and ray-plane intersection calculation. This dataset aims to tackle the challenge of inaccurate depth perception of glass surfaces in robotic navigation, and serves as a standardized evaluation benchmark for depth completion and scene reconstruction.
GlassRecon数据集概述
数据集简介
GlassRecon是一个包含室内场景中详细标注的玻璃深度图的数据集。
当前状态
数据集及相关代码即将发布。




