Get3DHuman: Lifting StyleGAN-Human into a 3D Generative Model using Pixel-aligned Reconstruction Priors (ICCV 2023)

Project Page | Paper | Code | Data

This is the official PyTorch implementation of Get3DHuman.

TODO:

• Synthetic data (with latent code)
• Inference code
• Pretrained weights
• Training Code

Requirements:

• Python 3
• PyTorch tested on 1.8.0+cu111

Inference:

• Download pretrained models from the following link and copy them into a same file.

S&T pretrained model or S&T pretrained model_v2.

Re pretrained model

• Enter the code path and run:

cd GET3DHUMAN_CODE_PATH
pip install -r requirements.txt
python inference.py --model_path PRETRAINED_MODELS_PATH --sample_num 1  

The results will be saved in "./results". Like sample

Note: A GTX 3090 is recommended to run Get3DHuman, make sure enough GPU memory if using other cards.

Overview of our framework

Multi-view images rendered by Blender.

Applications

Interpolation

Re-texturing

Inversion

Rendering methods

Citation

If you use Get3DHuman in your research, please consider the following BibTeX entry and give a star🌟!

@inproceedings{xiong2023Get3DHuman,
    author    = {Xiong, Zhangyang and Kang, Di and Jin, Derong and Chen, Weikai and Bao, Linchao and Cui, Shuguang and Han, Xiaoguang},
    title     = {Get3DHuman: Lifting StyleGAN-Human into a 3D Generative Model Using Pixel-Aligned Reconstruction Priors},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {9287-9297}
}

Acknowledgements

Here are some great resources we benefit or utilize from:

• Stylegan-Human
• PIFu
• Stylegan2