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] [Paper] [Slides] [Tensorflow Code] [Pytorch Code]

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 and Matthias Muller.

Requirements

Pytorch>=1.2.0
pytorch_geometric>=1.3.0
tensorflow graphics only used for tensorboard visualization

Install enviroment by runing:

source deepgcn_env_install.sh

Code Architecture

.
├── misc                    # Misc images
├── utils                   # Common useful modules
├── gcn_lib                 # gcn library
│   ├── dense               # gcn library for dense data (B x C x N x 1)
│   └── sparse              # gcn library for sparse data (N x C)
├── examples 
│   ├── sem_seg_dense       # code for point clouds semantic segmentation on S3DIS (data type: dense)
│   ├── sem_seg_sparse      # code for point clouds semantic segmentation on S3DIS (data type: sparse)
│   ├── part_sem_seg        # code for part segmentation on PartNet
│   └── ppi                 # code for node classification on PPI dataset
└── ...

How to train, test and evaluate our models

Please look the details in Readme.md of each task inside examples folder. All the information of code, data, and pretrained models can be found there.

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

Thanks for Guocheng Qian for the implementation of the Pytorch version.

rxlgq/deep_gcns_torch