Download dataset

Name Size (m2) Lidar scan [raw / merged / merged & resampled] RGB-D
sequence
Our fragments Our reconstruction Our camera poses
Apartment 69.17 (5.7GB) / (1.3GB) / (0.5GB) (5.5GB) (0.8GB) (129MB) (5.6MB)
Bedroom 21.01 (2.6GB) / (0.8GB) / (0.2GB) (3.9GB) (0.5GB) (80MB) (3.9MB)
Boardroom 60.90 (4.1GB) / (1.1GB) / (0.4GB) (4.4GB) (0.8GB) (139MB) (4.3MB)
Lobby 86.46 (3.3GB) / (1.5GB) / (0.5GB) (3.7GB) (0.7GB) (208MB) (3.5MB)
Loft 34.74 (2.8GB) / (1.1GB) / (0.3GB) (5.0GB) (0.8GB) (100MB) (4.5MB)

    Lidar scan

Lidar scan data was collected using a FARO Focus 3D X330 HDR scanner. The scanner has an operating range of 0.6m to 330m. At a distance of 10 meters, its ranging accuracy is 0.1 millimeters. Each scene was scanned from multiple locations. The scans were merged using dedicated software provided by the manufacturer, which is used for range scan alignment in industrial applications. It is in PLY file format.

    RGB-D sequence

Continuous RGB-D facetime-video sequences are captured using an Asus Xtion Live camera. The lengths of the sequences range from 11 to 18 minutes. Each sequence thoroughly covers the respective scene. The RGB-D sequences can be used as input to any scene reconstruction system.

Each RGB-D Sequence is packaged in a zip archive that contains consecutive color images stored as JPG and depth images stored as 16-bit PNG, where pixel values represent depth in millimeters. The resolution is 640×480, the frame rate is 30Hz. Color images and depth images are synchronized using software trigger. The first part of a file name indicates the synchronized frame number and the second part provides a time stamp in microseconds. The focal length is 525 for both axes and the principal point is (319.5, 239.5). The depth images are registered to the color images.

    Our reconstruction

The results from our scene reconstruction system. Please check the paper for techincal details and comparison to other methods. It is written in ply format containg meshed surface. It is in PLY file format.

    Our camera poses

Estimated 6-DoF camera poses for the every frames. It is written in log file format or open3d.posegraph json format.