Awesome

Gaussian GAN Decoder

Accepted at CVPRW!

Official implementation of: https://arxiv.org/abs/2404.10625

Project page: https://florian-barthel.github.io/gaussian_decoder/index.html

In this work, we present a novel approach that combines the high rendering quality of NeRF-based 3D-aware GANs with the flexibility and computational advantages of 3DGS. By training a decoder that maps implicit NeRF representations to explicit 3D Gaussian Splatting attributes, we can integrate the representational diversity and quality of 3D GANs into the ecosystem of 3D Gaussian Splatting for the first time. Additionally, our approach allows for a high resolution GAN inversion and real-time GAN editing with 3D Gaussian Splatting scenes.

Installation

# clone
git clone https://github.com/fraunhoferhhi/gaussian_gan_decoder.git --recursive
cd gaussian_gan_decoder

# conda environment
conda env create -f environment.yml
conda activate gs-decoder

Download the face ID weights from here and place the whole folder under: main/pretrained_models

Download Pre-trained GAN Models

Start Training

cd main
python train_pano2gaussian_decoder.py

You can find all training parameters as click.commands in train_pano2gaussian_decoder.py By default PanoHead is decoded.

Visualize Results

Download Pre-trained Gaussian GAN Decoder Models from here.

Our results can be rendered with our custom build viewer splatviz. Install the environment as described in the repository of the viewer and add the following three pip dependencies:

• scikit-image
• trimesh
• ninja

In order to load the GAN, you will also have to append the sys paths of this repository to the run_main.py like this:

Then, launch the run_main.py in the splatviz repository with the --mode=decoder flag as follows:

run_main.py --data_path=/folder/with/pkl/files --mode=decoder --ggd_path=/home/projects/gaussian_gan_decoder

If the head is rendered upside down, flip the Up-Vector in the Camera tab. Use the latent widget to switch between IDs. Using the gaussian_viewer, you can also export .ply files of the current scene.

Cite

@misc{barthel2024gaussian,
    title={Gaussian Splatting Decoder for 3D-aware Generative Adversarial Networks}, 
    author={Florian Barthel and Arian Beckmann and Wieland Morgenstern and Anna Hilsmann and Peter Eisert},
    year={2024},
    eprint={2404.10625},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

Acknowledgements

This research has partly been funded by the German Research Foundation (3DIL, grant no. 502864329), the European Union's Horizon Europe research and innovation programme (Luminous, grant no. 101135724), and the German Ministry of Education and Research (MoDL, grant no. 01IS20044).