PyTorch implementation of 3D-VGT(3D-VAE-GAN-Transformer)

This repository contains the source code for the paper “3D reconstruction method based on a generative model in continuous latent space”

Author email: [email protected]

Github Github


We use the renderings of ShapeNet in our experiments,which are available below:

Model Structure

train stage


Clone the Code Repository

git clone

Install Python Denpendencies

cd 3D-VGT
pip install -r 3D-VGT_requirements.txt

Update Settings in of part1

You need to update hyperparametersthe of the model and path of the datasets :

    parser.add_argument('--img_dir', type=str, default='../dataset/shapenet/train_imgs', help='input images path')
    parser.add_argument('--vox_dir', type=str, default='../dataset/shapenet/train_voxels', help='input voxels path')
    parser.add_argument('--lr', type=float, default='0.0002', help='learning rate')
    parser.add_argument('--batch_size', type=int, default='32', help='batch_size in training')
    parser.add_argument("--epoch", type=int, default=500, help="epoch in training")

Part of the experimental results are as follows

results of ShapeNet:

Get Started

To train UAGAN, you can simply use the following command:

stage 1

cd 3D-VGT
cd part1

stage 2

cd 3D-VGT
cd part2

Note: Since our paper has not been published yet, we only show the model structure and training code. When the paper is published, we will publish the full work of the paper. Welcome scholars to discuss and exchange.


This project is open sourced under MIT license.


View Github