Structured3D
收藏资源简介:
Structured3D是由上海科技大学等机构创建的大型合成数据集,包含21,835个房间的超过196,515张照片级真实感图像,旨在为结构化3D建模任务提供丰富的3D结构标注。该数据集利用专业室内设计资源,通过先进的渲染引擎自动提取和生成高质量的3D结构信息。创建过程中,数据集不仅捕捉了多种3D结构及其相互关系,还通过模拟真实光照条件增强了图像的真实感。Structured3D的应用领域广泛,主要用于训练深度网络进行房间布局估计,解决从2D数据推断3D信息的核心计算机视觉问题。
Structured3D is a large-scale synthetic dataset developed by ShanghaiTech University and other affiliated institutions. It comprises over 196,515 photorealistic images across 21,835 individual rooms, and is intended to provide abundant 3D structural annotations for structured 3D modeling tasks. Leveraging professional indoor design resources, the dataset automatically extracts and generates high-quality 3D structural information via advanced rendering engines. During its construction, the dataset not only captures diverse 3D structures and their interrelations but also enhances image realism by simulating real-world lighting conditions. Structured3D has a wide range of application domains, primarily used for training deep networks to perform room layout estimation, addressing the core computer vision problem of inferring 3D information from 2D data.




