/GCN_Final_DeepGCNs

Course GCN final project

Primary LanguagePythonMIT LicenseMIT

GCN_Final_DeepGCNs

Final project of the course Graph Convolutional Networks.
This repository is partly borrowed from original paper's code, and I fixed part of code for running without problem.
This repository is only for the final porject.

DeepGCNs: Can GCNs Go as Deep as CNNs?

Requirements

There are many package version requirements, so please install a new conda enviroment to run the code.

Install the conda enviroment by running:

conda env create -f deepgcn.yml
conda activate deepgcn

Code Architecture

.
├── images                  # images
├── utils                   # common useful modules
├── gcn_impl                # gcn library
├── sem_seg                 # code for point clouds semantic segmentation on S3DIS 
└── ...

How to train, test and evaluate the models (important!)

Please look the details in README.md in sem_seg folder.
All the information of the code, data, and pretrained models can be found there.

A simple example (training from scratch)

First enter the example directory,

cd sem_seg/

and then run:

CUDA_VISIBEL_DIVICES=0,1,2,3 python train.py --phase train --multi_gpus --batch_size 8

Lower the batch size if out of memory. The batch size will not influence the test results.

Contact

Tairen Piao

E-mail: piaotairen@snu.ac.kr