megayeye / scenecad

[ECCV 2020] SceneCAD: Predicting Object Alignments and Layouts in RGB-D Scans

SceneCAD (ECCV 2020)

We digitize 3D scans by jointly estimating scene layout and predicting CAD model alignments

Download Paper (.pdf)

See Youtube Video

Download Data

Get the SceneCAD layout dataset here!

Please cite our work when using our dataset :)

Description

We introduce a new richly-annotated real-world scene layout dataset con-sisting of 1151 CAD shells and wireframes on top of the ScanNet RGB-Ddataset, allowing large-scale data-driven training for layout estimation.

For the public dataset, we provide annotations with:

• 13756 Corners
• 20509 Edges
• 8376 Planes
• 1151 real-world 3D scans

Our annotations are:

• Non-Manhattan Rooms
• Non-Convex Shells
• n-Polygonal Planes

Lightweight Layout Prediction Method

Pytorch code coming soon!

Sample Annotations

Citation

If you use this dataset or code please cite:

@InProceedings{avetisyan2020scenecad,
    title={SceneCAD: Predicting Object Alignments and Layouts in RGB-D Scans},
    author={Armen Avetisyan and Tatiana Khanova and Christopher Choy and Denver Dash and Angela Dai and Matthias Nießner},
    booktitle = {The European Conference on Computer Vision (ECCV)},
    month={August},
    year={2020},
}