Core Insights - Realsee has launched Realsee3D, a dataset containing 10,000 indoor 3D scenes, aimed at supporting academic research and non-commercial purposes, potentially making it the largest spatial 3D dataset globally [1][3] - The dataset is designed to enhance research in indoor 3D perception, reconstruction, and scene understanding, addressing the existing gap in high-quality spatial data [3][6] Dataset Features - The Realsee3D dataset includes 10,000 unique indoor scenes, comprising 95,962 rooms and 299,073 viewpoints/RGB-D pairs, providing a rich resource for researchers [8] - It features panoramic RGB-D captures with complete room-level coverage and comprehensive annotations, including CAD drawings, floor plans, and semantic segmentation [8] - The dataset consists of 1,000 real-world scenes with diverse layouts and decoration styles, along with 9,000 procedurally generated scenes using over 100 designer-curated style templates [8] Research Applications - The dataset is suitable for core research directions in spatial intelligence, such as geometric reconstruction, multi-modal learning, and embodied AI, encouraging global researchers and developers to explore the future of spatial intelligence [6][7]

此前，如视发布空间深度推测大模型 Argus1.0。作为首个支持全景图输入的空间深度推测大模型，Argus1.0依托如视所积累的千万级三维空间数据库训练 而成。而此次开放的 Realsee3D三维数据集，正是从这一海量数据库中精选出的高质量样本。 数据集优势 Realsee3D 是一个大规模多视角RGB-D数据集，旨在推动室内三维感知、重建与场景理解领域的研究。该数据集具有以下优势特征： 2025年12月16日，如视宣布，面向学术研究及非商业用途正式开放10000套室内三维数据集 Realsee3D——这或是全球目前最大规模的空间三维数据集， 旨在为空间智能领域的研究者、开发者提供高质量数据基础，加速整个行业的技术迭代与应用落地。 Realsee3D 场景多样 为了确保模型在复杂现实环境中的鲁棒性，我们采用了"真实数据+程序化生成"的双引擎策略： 数据类型 - 10,000个独特的室内三维场景 - 95,962个细分房间单元 - 299,073组视点/RGB-D 图像对 - 几何层：提供高精度 CAD 图纸与平面图。 - 语义层：包含 2D语义分割与3D检测标签。 - 采集数据：1,000 个真实场景（捕捉现 ...