By Jun Li, Kai Xu, Siddhartha Chaudhuri, Ersin Yumer, Hao Zhang, Leonidas Guibas
This repository contains the pre-trained models for box structure generation, as well as the training/testing code for the generation model.
Details of the work can be found here.
If you find our work useful in your research, please consider citing:
@article {li_sig17,
title = {GRASS: Generative Recursive Autoencoders for Shape Structures},
author = {Jun Li and Kai Xu and Siddhartha Chaudhuri and Ersin Yumer and Hao Zhang and Leonidas Guibas},
journal = {ACM Transactions on Graphics (Proc. of SIGGRAPH 2017)},
volume = {36},
number = {4},
pages = {to appear},
year = {2017}
}
Training
Run trainTestVaeGan.m to train the vae-gan model on the provided chair dataset.
Testing
Use test_demo.m to generate shapes based on trained model. There is already a pre-trained model inside. The generated shape structures could be visulized in matlab.
For any questions, please contact Jun Li(jun.johnson.li@gmail.com) and Kai Xu(kevin.kai.xu@gmail.com).