SnapNet: Unstructured point cloud semantic labeling using deep segmentation networks

Published in Computer and Graphics, 2017

Authors: Alexandre Boulch, Bertrand Le Saux and Nicolas Audebert

Code available at Github

Computer and Graphics version

Link to the paper.

@article{boulch2017snapnet,
  title={SnapNet: 3D point cloud semantic labeling with 2D deep segmentation networks},
  author={Boulch, Alexandre and Guerry, Joris and Le Saux, Bertrand and Audebert, Nicolas},
  journal={Computers \& Graphics},
  year={2017},
  publisher={Elsevier}
}

3DOR Version

A preliminary version of the this work has been presented at 10th Eurographics workshop on 3D Object retrieval, 3DOR 2017.

Download paper here

@inproceedings{boulch2017unstructured,
  title={Unstructured point cloud semantic labeling using deep segmentation networks},
  author={Boulch, Alexandre and Saux, Bertrand Le and Audebert, Nicolas},
  booktitle={Eurographics Workshop on 3D Object Retrieval},
  volume={2},
  year={2017}
}

CODE

The code is available in the dedicated Github repository.

The network weights are available to download:

RGB model

Composite model

Fusion model (to be used with RGB and Composite)

NPY model of VGG for initialization (based on caffe model VGG 16 from caffe model zoo, converted with Kaffe)