Home

Awesome

GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images (NeurIPS 2022)<br><sub>Official PyTorch implementation </sub>

Teaser image

GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images<br> Jun Gao , Tianchang Shen , Zian Wang, Wenzheng Chen, Kangxue Yin , Daiqing Li, Or Litany, Zan Gojcic, Sanja Fidler <br> Paper , Project Page

Abstract: As several industries are moving towards modeling massive 3D virtual worlds, the need for content creation tools that can scale in terms of the quantity, quality, and diversity of 3D content is becoming evident. In our work, we aim to train performant 3D generative models that synthesize textured meshes which can be directly consumed by 3D rendering engines, thus immediately usable in downstream applications. Prior works on 3D generative modeling either lack geometric details, are limited in the mesh topology they can produce, typically do not support textures, or utilize neural renderers in the synthesis process, which makes their use in common 3D software non-trivial. In this work, we introduce GET3D, a Generative model that directly generates Explicit Textured 3D meshes with complex topology, rich geometric details, and high fidelity textures. We bridge recent success in the differentiable surface modeling, differentiable rendering as well as 2D Generative Adversarial Networks to train our model from 2D image collections. GET3D is able to generate high-quality 3D textured meshes, ranging from cars, chairs, animals, motorbikes and human characters to buildings, achieving significant improvements over previous methods.

Teaser Results

For business inquiries, please visit our website and submit the form: NVIDIA Research Licensing

News

Requirements

Server usage through Docker

cd docker
chmod +x make_image.sh
./make_image.sh get3d:v1

Preparing datasets

GET3D is trained on synthetic dataset. We provide rendering scripts for Shapenet. Please refer to readme to download shapenet dataset and render it.

Employing FlexiCubes

We integrate FlexiCubes, our lastest high-quality isosurface representation. To leverage FlexiCubes as an alternative to DMTet for isosurfacing, simply append --iso_surface flexicubes to the following training and inference commands.

Train the model

Clone the gitlab code and necessary files:

cd YOUR_CODE_PATH
git clone git@github.com:nv-tlabs/GET3D.git
cd GET3D; mkdir cache; cd cache
wget https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/metrics/inception-2015-12-05.pkl

Train the model

cd YOUR_CODE_PATH 
export PYTHONPATH=$PWD:$PYTHONPATH
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=40 --data_camera_mode shapenet_car  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=80 --data_camera_mode shapenet_motorbike  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=400 --data_camera_mode shapenet_chair  --dmtet_scale 0.8  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=40 --data_camera_mode shapenet_car  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 0
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=80 --data_camera_mode shapenet_motorbike  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 0
python train_3d.py --outdir=PATH_TO_LOG --data=PATH_TO_RENDER_IMG --camera_path PATH_TO_RENDER_CAMERA --gpus=8 --batch=32 --gamma=3200 --data_camera_mode shapenet_chair  --dmtet_scale 0.8  --use_shapenet_split 1  --one_3d_generator 0

If want to debug the model first, reduce the number of gpus to 1 and batch size to 4 via:

--gpus=1 --batch=4

Inference

Inference on a pretrained model for visualization

python train_3d.py --outdir=save_inference_results/shapenet_car  --gpus=1 --batch=4 --gamma=40 --data_camera_mode shapenet_car  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0 --inference_vis 1 --resume_pretrain MODEL_PATH
python train_3d.py --outdir=save_inference_results/shapenet_chair  --gpus=1 --batch=4 --gamma=40 --data_camera_mode shapenet_chair  --dmtet_scale 0.8  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0 --inference_vis 1 --resume_pretrain MODEL_PATH
python train_3d.py --outdir=save_inference_results/shapenet_motorbike  --gpus=1 --batch=4 --gamma=40 --data_camera_mode shapenet_motorbike  --dmtet_scale 1.0  --use_shapenet_split 1  --one_3d_generator 1  --fp32 0 --inference_vis 1 --resume_pretrain MODEL_PATH

Evaluation metrics

Compute FID
Compute COV & MMD scores for LFD & CD

License

Copyright © 2022, NVIDIA Corporation & affiliates. All rights reserved.

This work is made available under the Nvidia Source Code License .

Broader Information

GET3D builds upon several previous works:

Citation

@inproceedings{gao2022get3d,
title={GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images},
author={Jun Gao and Tianchang Shen and Zian Wang and Wenzheng Chen and Kangxue Yin
and Daiqing Li and Or Litany and Zan Gojcic and Sanja Fidler},
booktitle={Advances In Neural Information Processing Systems},
year={2022}
}