The Largest-Scale Globally: Realsee Open-Sources Indoor 3D Dataset Realsee3D

BEIJING, Dec. 17, 2025 /PRNewswire/ -- Realsee announced the official opening of Realsee3D, a dataset of 10,000 indoor 3D scenes, for academic research and non-commercial purposes.

This is potentially the world's largest spatial 3D dataset to date, aiming to provide a high-quality data foundation for researchers and developers in the field of spatial intelligence, and accelerate the technological iteration and application implementation of the entire industry. 

Prior to this, Realsee released Argus 1.0, a large model for spatial depth estimation. As the first large model for spatial depth estimation that supports panoramic image input, Argus 1.0 was trained based on the 10-million-level 3D spatial database accumulated by Realsee. The Realsee3D 3D dataset opened this time is precisely a selection of high-quality samples from this massive database. 

Overview

Realsee3D is a large-scale multi-view RGB-D dataset designed to advance research in indoor 3D perception, reconstruction, and scene understanding.

Features

• Large Scale: 10,000 unique indoor scenes, comprising 95,962 rooms and 299,073 viewpoints/RGB-D pairs.
• Rich Data: Panoramic RGB-D captures with complete room-level coverage.
• Comprehensive Annotations: Includes CAD drawings, floor plans, semantic segmentation, 3D detection labels and more (forthcoming).
• Diverse Scenes: Comprising 1,000 real-world scenes with varied layouts and decoration styles, and 9,000 procedurally generated scenes utilizing over 100 designer-curated style templates, ensuring diverse furniture models and styles for robust training and testing.

Data Types

• HDR RGB Panoramic
• Depth Map
• Poses
• CAD Drawings
• Floorplans
• Segmentation
• 3D object detection information

Applicable Research Directions

For a long time, research and applications in the field of spatial intelligence have faced a bottleneck challenge: there has always been a huge gap in high-quality spatial data. Relying on its technological accumulation and resource reserves in the field of 3D spatial data, Realsee is filling this gap. 

This dataset is suitable for core research directions in spatial intelligence such as geometric reconstruction, multi-modal learning, and embodied AI. Researchers and developers worldwide are welcome to download and use the Realsee3D 3D dataset to jointly explore the future boundaries of spatial intelligence research. 

Data Organization & Access

Currently, the Realsee3D dataset is open for application through official channels. Visit the Realsee GitHub repository to access it: 

https://github.com/realsee-developer/RealSee3D

SOURCE Beike Realsee Technology (HK) Limited