voidrank / Geo-CNN

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Modeling Local Geometric Structure of 3D Point Clouds using Geo-CNN

Bibtex

@InProceedings{Lan_2019_CVPR,
    author = {Lan, Shiyi and Yu, Ruichi and Yu, Gang and Davis, Larry S.},
    title = {Modeling Local Geometric Structure of 3D Point Clouds Using Geo-CNN},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year = {2019}
}

Installation and Usage

We re-implemented the Geo-CNN following Frustum PointNets.

Follow the instruction of installing Frustum PointNets and thus compile Geo-Conv operator located at models/tf_ops/geoconv.

Use scripts/command_train_geocnn_v1.sh and command_test_geocnn_v1.sh to train/test Geo-CNN.

TODO

  • Combine GeoCNN and PointNet++
  • GeoCNN on other 3D datasets (ModelNet40, ScanNet)

Others

  • Well-trained parameters
  • This implementation is slightly different from the original version on a private deep learning architure.

About

License:Apache License 2.0


Languages

Language:Python 69.9%Language:C++ 19.9%Language:Cuda 8.6%Language:Shell 1.6%