Supplemental code and scripts for the paper SceneNN: A Scene Meshes Dataset with aNNotations
C++ Python Makefile
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


SceneNN is an RGB-D dataset with more than 100 indoor scenes. All scenes are carefully annotated with instance-level segmentation.

Please refer to our project page at for the papers and scene previews in the supplementary document.

To access the dataset, please use following repositories on Google Drive:

1. Instance-level segmentation

This includes all 90+ scenes captured by Asus Xtion PRO.

Please use the Python script scenenn_download to download a few specific scenes or the entire dataset. To run this script, please install PyDrive.

3D annotation

Each scene is identified by an ID with three digits, e.g., 076, 123. The data of the scene is structured as follows:

├── 123
│   ├── 123.ply                 /* the reconstructed triangle mesh  */
│   ├── 123.xml                 /* the annotation                   */
│   ├── trajectory.log          /* camera pose (local to world)     */
└── oni
    └── 123.oni                 /* the raw RGBD video               */
└── intrinsic
    └── asus.ini                /* intrinsic matrix data for Asus Xtion camera  */
	└── kinect2.ini             /* intrinsic matrix data for Kinect v2 camera   */

Depending on which application you are working on, you might want to download only the mesh with annotation (~100 - 200MB per scene), or include the raw RGBD (~1 GB per scene). The object tag and axis-aligned bounding box are stored separately in the XML file. If you need object poses, please find them in the beta repository, below.

Color meshes

The color meshes for each scene could be downloaded from our experimental repository here: []

2D annotation

Please download them here:

We encode the label to color channels in RGB format. To decode, please use this simple C++ code.

Raw RGB-D videos

We also provide raw RGB-D files that are output from the Asus Xtion camera. They are stored in the oni folder in both repositories. (You might need to scroll down to the bottom of the scene list to see it.)

In order to extract depth and color images from ONI files, please use the tool in the playback folder from this repository. After extraction, depth map is stored in 16-bit unsigned short format. The depth unit is in millimeter

For Windows, a precompiled binary is provided. Note that OpenNI 1.5.4 x64 must be installed.

For Linux, a MakeFile is included in the playback folder. The compilation is tested on Ubuntu 16. (Credit and great thanks to Joao Henriques (joao [at] for this handy script!)

sudo apt-get install libopenni-dev libfreeimage-dev libglew-dev freeglut3-dev
make all

Camera pose

The trajectory.log contains blocks of 5 lines:

  • The first line in each block indicates the frame index.
  • The next 4 lines is a 4x4 matrix. This matrix transforms a point from camera space to world space.

Intrinsic matrix is in the format of [fx, 0, cx; 0, fy, cy; 0 0 1], where fx, fy, cx, cy could be found in asus.ini or kinect2.ini, respectively.

2. Category-level segmentation (beta)

This includes our experimental category-level segmentation and object pose annotation:

  • Semantic labels using the 40 classes defined in NYU-D v2.
  • Object poses and better floor aligned meshes.

Such annotation will be merged to the main repository once they are stable.

This repository also includes scenes captured by both Asus Xtion PRO and Kinect v2.


ObjectNN is a subset of SceneNN that has all RGB-D objects extracted and split into 20 categories. We used it for our SHREC 2017 track: RGB-D to CAD object retrieval. For more information, please see here.


Our datasets are free for educational and research use. We would love to hear from you if you find them useful. If you use the data, please cite

    author = {Binh-Son Hua and Quang-Hieu Pham and Duc Thanh Nguyen and Minh-Khoi Tran and Lap-Fai Yu and Sai-Kit Yeung},
    title = {SceneNN: A Scene Meshes Dataset with aNNotations},
    booktitle = {International Conference on 3D Vision (3DV)},
    year = {2016}

If you intend to use the data for commercial applications, please kindly contact us beforehand at scenenn[at]