GCN for 3D Point Cloud Classification

This repository is a toy PyTorch implementation of GCN-based 3D point cloud classification model.

Requirements

• Python3==3.7
• pytorch==1.4.0
• tensorboardX==2.0
• hdf5==1.10.4

Dataset

To evaluate the model, ModelNet40 dataset in HDF5 format are required to be downloaded and unzipped to the data folder.

Download ModelNet40 dataset for classification task by running the following commands:

cd ./data
wget https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip
unzip modelnet40_ply_hdf5_2048.zip
rm modelnet40_ply_hdf5_2048.zip

Usage

To train the model, run

python main.py --phase train --device 0 --train-batch-size 32

The log files, network parameters, and TensorBoard logs will be saved to results folder by default. We can use TensorBoard to view the training progress:

tensorboard --logdir ./results/tensorboard

For more hyper-parameters, please refer to point_gcn/tools/configuration.py.

To evaluate the model, run

python main.py --phase test --device 0 --test-batch-size 32 --weights [checkpoints] 

You should specify the [checkpoints]. For instance:

python main.py --phase test --device 0 --test-batch-size 32 --weights ./results/models/model1.pt

Acknowledgement

Our code is released under MIT License (see LICENSE for details).