https://github.com/lightaime/deep_gcns

Tensorflow Repo for "DeepGCNs: Can GCNs Go as Deep as CNNs?" ICCV2019 Oral https://www.deepgcns.org
https://github.com/lightaime/deep_gcns

3d-point-clouds deep-gcns geometric-deep-learning graph-neural-networks

Last synced: 3 months ago
JSON representation

Tensorflow Repo for "DeepGCNs: Can GCNs Go as Deep as CNNs?" ICCV2019 Oral https://www.deepgcns.org

• Host: GitHub
• URL: https://github.com/lightaime/deep_gcns
• Owner: lightaime
• License: mit
• Created: 2019-04-05T23:52:51.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-01-23T13:17:32.000Z (over 4 years ago)
• Last Synced: 2024-01-16T19:12:58.173Z (5 months ago)
• Topics: 3d-point-clouds, deep-gcns, geometric-deep-learning, graph-neural-networks
• Language: Python
• Homepage:
• Size: 219 MB
• Stars: 622
• Watchers: 20
• Forks: 89
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Lists

• awesome-gcn - lightaime/deep_gcns
• awesome-stars - lightaime/deep_gcns - Tensorflow Repo for "DeepGCNs: Can GCNs Go as Deep as CNNs?" ICCV2019 Oral https://www.deepgcns.org (Python)

README

        
# DeepGCNs: Can GCNs Go as Deep as CNNs?

In this work, we present new ways to successfully train very deep GCNs. We borrow concepts from CNNs, mainly residual/dense connections and dilated convolutions, and adapt them to GCN architectures. Through extensive experiments, we show the positive effect of these deep GCN frameworks.

[[Project]](https://www.deepgcns.org/) [[Paper]](https://arxiv.org/abs/1904.03751) [[Slides]](https://docs.google.com/presentation/d/1L82wWymMnHyYJk3xUKvteEWD5fX0jVRbCbI65Cxxku0/edit?usp=sharing) [[Tensorflow Code]](https://github.com/lightaime/deep_gcns) [[Pytorch Code]](https://github.com/lightaime/deep_gcns_torch)





## Overview

We do extensive experiments to show how different components (#Layers, #Filters, #Nearest Neighbors, Dilation, etc.) effect `DeepGCNs`. We also provide ablation studies on different type of Deep GCNs (MRGCN, EdgeConv, GraphSage and GIN).





Further information and details please contact [Guohao Li](https://ghli.org) and [Matthias Müller](https://matthias.pw/).

## Requirements

* [TensorFlow 1.12.0](https://www.tensorflow.org/)

* [h5py](https://www.h5py.org/)

* [vtk](https://vtk.org/) (only needed for visualization)

* [jupyter notebook](https://jupyter.org/) (only needed for visualization)

## Conda Environment

In order to setup a conda environment with all neccessary dependencies run,

```

conda env create -f environment.yml

```

## Getting Started

You will find detailed instructions how to use our code for semantic segmentation of 3D point clouds, in the folder [sem_seg](sem_seg/). Currently, we provide the following:

* Conda environment

* Setup of S3DIS Dataset

* Training code

* Evaluation code

* Several pretrained models

* Visualization code

## Citation

Please cite our paper if you find anything helpful,

```

@InProceedings{li2019deepgcns,

    title={DeepGCNs: Can GCNs Go as Deep as CNNs?},

    author={Guohao Li and Matthias Müller and Ali Thabet and Bernard Ghanem},

    booktitle={The IEEE International Conference on Computer Vision (ICCV)},

    year={2019}

}

```

```

@misc{li2019deepgcns_journal,

    title={DeepGCNs: Making GCNs Go as Deep as CNNs},

    author={Guohao Li and Matthias Müller and Guocheng Qian and Itzel C. Delgadillo and Abdulellah Abualshour and Ali Thabet and Bernard Ghanem},

    year={2019},

    eprint={1910.06849},

    archivePrefix={arXiv},

    primaryClass={cs.CV}

}

```

## License

MIT License

## Acknowledgement

This code is heavily borrowed from [PointNet](https://github.com/charlesq34/pointnet) and [EdgeConv](https://github.com/WangYueFt/dgcnn). We would also like to thank [3d-semantic-segmentation](https://github.com/VisualComputingInstitute/3d-semantic-segmentation) for the visualization code.