Matterport3D is a pretty awesome dataset for RGB-D machine learning tasks :)
C++ C Lua Objective-C CMake Makefile
Switch branches/tags
Nothing to show
Clone or download
Failed to load latest commit information.
code Update Apr 4, 2018
docs add arxiv pdf link to index.html Sep 20, 2017
img image Sep 20, 2017
metadata fix link to mpcat40.tsv Sep 20, 2017
tasks benchmark split Oct 22, 2017
LICENSE.txt Create LICENSE.txt Sep 21, 2017 Update Sep 22, 2017 Update Aug 9, 2018



The Matterport3D V1.0 dataset contains data captured throughout 90 properties with a Matterport Pro Camera.

This repository includes the raw data for the dataset plus derived data, annotated data, and scripts/models for several scene understanding tasks.

Visit the main website for updates and to browse the data.


Matterport3D: Learning from RGB-D Data in Indoor Environments

If you use the Matterport3D data or code please cite:

  title={{Matterport3D}: Learning from {RGB-D} Data in Indoor Environments},
  author={Chang, Angel and Dai, Angela and Funkhouser, Thomas and Halber, Maciej and Niessner, Matthias and Savva, Manolis and Song, Shuran and Zeng, Andy and Zhang, Yinda},
  journal={International Conference on 3D Vision (3DV)},


The dataset consists of several types of annotations: color and depth images, camera poses, textured 3D meshes, building floor plans and region annotations, object instance semantic annotations. For details see the data organization document.

To download the dataset, you must indicate that you agree to the terms of use by signing the Terms of Use agreement form and sending it to: We will then provide download access to the dataset.

Benchmark tasks

Using the Matterport3D data, we present several benchmark tasks: image keypoint matching, view overlap prediction, surface normal estimation, region type classification, and semantic voxel labeling. See the tasks directory for details.


We provide code for loading and viewing the data. See the code directory for details.


The data is released under the Matterport3D Terms of Use, and the code is released under the MIT license.