This repository contains the reproduce codes for the paper Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling.
here use Matlab code to process ShapeNetCore/ModelNet10/ModelNet40 dataset, convert .off type to voxel type and save it in .mat file
3d-gan network is being build and train using tensorflow r-0.12
run the python file. It will start build the 3D-GAN network and start training
load_data.py contains two functions: load_data_np, load_data_path, which is used to load .mat file and generate train files list
3dgan_model.py contains class GAN_3D, which can build the network and do training. this file is modified based on DCGAN in Tensorflow, by changing input and filters size and dim.
For some reasons, this repository is unfinished and if you are interesting in it, please contact me and work together