Yunji Seo*, Young Sun Choi*, Hyun Seung Son, Youngjung Uh

We introduce integrating a Flexible Level of Detail (FLoD) to 3DGS, to allow a scene to be rendered at varying levels of detail according to hardware capabilities.

Our code was tested on conda environment installed with environment.yml and the submodules below.

git clone https://github.com/3DGS-FLoD/flod.git
cd flod

Setup conda environment

conda env create -f environment.yml
conda activate flod

Clone submodules

mkdir submodules
git clone https://github.com/graphdeco-inria/diff-gaussian-rasterization submodules/diff-gaussian-rasterization
git clone https://gitlab.inria.fr/bkerbl/simple-knn.git submodules/simple-knn

Install dependencies

sudo apt install libglm-dev
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn

To reproduce, run...

train_{dataset_name}.sh # dl3dv / mipnerf / tnt

Render and evaluate by...

render_single.sh # for individual level rendering of 3DGS-FLoD
render_selective.sh # for selective rendering of 3DGS-FLoD

To run viewer as demonstrated on our project page

convert4viewer.sh
SIBR_viewers/install/bin/SIBR_flodViewer_app /path/to/your/model

We build our code for FLoD on top of the open-source code of 3D Gaussian Splatting.
Hence our licencse follows graphdeco-inria/gaussian-splatting

We would like to express our gratitude to the authors of the 3D Gaussian Splatting.
Their work has laid the foundation for this research.
Our code is largely based on their open-source project: graphdeco-inria/gaussian-splatting

@misc{seo2024flod,
      title={FLoD: Integrating Flexible Level of Detail into 3D Gaussian Splatting for Customizable Rendering}, 
      author={Yunji Seo and Young Sun Choi and Hyun Seung Son and Youngjung Uh},
      year={2024},
      eprint={2408.12894},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2408.12894}, 
}