Layout annotation on a subset of Matterport3D dataset
This dataset extends the Matterport3D dataset with general Manhattan layout annotations and is used in our work "Manhattan Room Layout Reconstruction from a Single 360° image: A Comparative Study of State-of-the-art Methods" for performance evaluation.
We released only the annotated layout files. To generate the corresponding RGB panoramas, use the following instructions:
{
"cameraHeight": 1.6, // distance between camera and floor
"layoutHeight": 2.9809624004364013, // distance between floor and ceiling
"layoutObj2ds": { // door or window labeling block
"num": 0,
"obj2ds": []
},
"layoutPoints": { // corner labeling block, we only annotate the corner in the horizontal direction because you can easily get the corner in the vertical direction from cameraHeight and layoutHeight.
"num": 6, // corner number in the horizontal direction
"points": [ // List of the corner
{
"coords": [ // uv coordintate of the corner on the panorama
0.7081447345651483, // u coordintate of the corner on the panorama
0.5 // v coordintate of the corner on the panorama. We annotate on horizon line, then calculate the vertical position by cameraHeight and layoutHeight.
],
"id": 0, // id number
"xyz": [ // project the corner on to the 3D coordinate
3.0078125, // x value
0.0, // y value
-0.8097623087756155 // z value
]
},
{...}
]
},
"layoutWalls": { // layout walls represented by plane equation
"num": 6, // wall number
"walls": [ // list of wall
{
"id": 0, // id number
"normal": [ // plane equation's normal
1.0,
0.0,
-0.0
],
"planeEquation": [ // plane equation
1.0, // a
0.0, // b
-0.0, // c
-3.0078125 // d
],
"pointsIdx": [ // which two points construct this wall
0,
1
],
"width": 2.8476272687756152 // the wall width
},
{...}
]
},
"panoId": "nothing" // a simple ID of panorama
}
<dir1>_<dir2>_label.json
follows the folder structure of Matterport3D dataset.If you use this dataset for your research, please cite our paper as follows:
@article{zou20193d,
title={Manhattan Room Layout Reconstruction from a Single 360$^{\circ}$ image: A Comparative Study of State-of-the-art Methods},
author={Zou, Chuhang and Su, Jheng-Wei and Peng, Chi-Han and Colburn, Alex and Shan, Qi and Wonka, Peter and Chu, Hung-Kuo and Hoiem, Derek},
journal={arXiv preprint arXiv:1910.04099},
year={2019}
}
The data is released under the Matterport3D Terms of Use, and the code is released under the MIT license. By using this dataset, you agree to all the terms in the Matterport3D Terms of Use.
For all the layouts in the dataset, we started with initial layouts from LayoutMP3D, which is also annotated with PanoAnnotator.