jsong-nj/graphcnn-gan

Graph-convolutional GAN for point cloud generation. Code from ICLR 2019 paper Learning Localized Generative Models for 3D Point Clouds via Graph Convolution

Learning Localized Generative Models for 3D Point Clouds via Graph Convolution (ICLR 2019)

Warning: trained models are large! A code-only repository is available at: https://github.com/diegovalsesia/GraphCNN-GAN-codeonly

If you like our work, please cite the journal version of the paper.

Journal version BibTex reference:

@ARTICLE{Valsesia2019journal,
author={Diego {Valsesia} and Giulia {Fracastoro} and Enrico {Magli}},
journal={IEEE Transactions on Multimedia},
title={Learning Localized Representations of Point Clouds with Graph-Convolutional Generative Adversarial Networks},
year={2019},
volume={},
number={},
pages={},
}

ICLR 2019 BibTex reference:

@inproceedings{valsesia2019learning,
  title={Learning Localized Generative Models for 3D Point Clouds via Graph Convolution},
  author={Valsesia, Diego and Fracastoro, Giulia and Magli, Enrico},
  booktitle={International Conference on Learning Representations (ICLR) 2019},
  year={2019}
}

Requirements

• Python 2.7
• Tensorflow >=1.6

Usage

A trained model for the method with aggregation upsampling is provided for the following Shapenet classes: airplane, chair, sofa, table.

• launch_test.sh : generate a batch of point clouds from the specified class
• launch_train.sh : retrain the network (requires downloading the Shapenet dataset and place it in the data directory)