splatwizard 0.0.4

Gaussian Splatting compression toolkit

Splatwizard: Gaussian Splatting Compression Toolkit

Splatwizard is a one-stop toolkit designed for researching 3DGS compression, dedicated to accelerating community exploration in this field. With flexible API design, you can easily combine advanced components from state-of-the-art models to build customized models. Key features of the project include:

• Easy-to-adapt evaluation framework for new models
• Comprehensive evaluation metrics
• Multiple baseline models included for comparison

Latest News

• 2025/11/25: Release first version.

Installation

Splatwizard can be installed via pip

pip install splatwizard

Since the installation process requires significant time to precompile all components, you can monitor the installation progress using the --verbose option.

pip install splatwizard --verbose

Additionally, pip compiles components in an isolated environment by default. You can use --no-build-isolation to perform the compilation directly in the current environment.

pip install splatwizard --verbose --no-build-isolation

Quick start

Train your first 3DGS model in splatwizard

sw-train \
    --source_path /data/MipNeRF-360/bicycle \
    --output_dir /output/gs \                  
    --model 3dgs \
    --optim 3dgs

Evaluate model

sw-eval \
    --source_path /data/MipNeRF-360/bicycle \
    --output_dir /output/gs \                  
    --model 3dgs \
    --optim 3dgs \
    --checkpoint /output/gs/checkpoints/ckpt30000.pth

For more models. please check Model List

Documentation

• Installation
• Training model
• How to develop your own model
• Model zoo

License

Splatwiard is licensed under the MIT License. The project incorporates code from other projects, which remains under their original licenses.