We reduce storage requirements of 3D Gaussian Splatting while using a minimal number of primitives. First, we determine the distinct Gaussian from the near ones, minimizing redundancy without sacrificing quality. Second, we propose a compact and precise attribute representation that efficiently captures both continuity and irregularity among primitives. Additionally, we propose a sub-vector quantization technique for improved irregularity representation, maintaining fast training with a negligible codebook size.

Our code is based on Mini-Splatting.

For installation: We recommend to use cuda 12.1 with python 3.11 for easy setup.

git clone https://github.com/maincold2/OMG.git
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

If you have trouble in installing cuml, please refer to the CUML Installation Guide.

We used Mip-NeRF 360, Tanks & Temples, and Deep Blending.

#for outdoor scenes (e.g., Mip-NeRF 360 outdoor and T&T scenes)
python train.py -s <path to COLMAP> -m <model path> --eval --imp_metric outdoor
#for indoor scenes (e.g., Mip-NeRF 360 indoor and DB scenes)
python train.py -s <path to COLMAP> -m <model path> --eval --imp_metric indoor

Threshold for importance-based scoring. 0.96, 0.98, 0.99, 0.999, 0.9999 for XS to XL. 0.96 (XS) by default

Both the ply file (3DGS format) and compressed file (comp.xz) are generated after traininig.

python render.py -s <path to COLMAP> -m <model path> --decode
python metrics.py -m <model path> 

Rendering with the compressed file (comp.xz), otherwise using the ply file. The results are the same regardless of this option.

@article{lee2025omg,
  author    = {Lee, Joo Chan and Ko, Jong Hwan and Park, Eunbyung},
  title     = {Optimized Minimal 3D Gaussian Splatting},
  journal   = {arXiv preprint arXiv:2503.16924},
  year      = {2025},
}