DDGS

This is the home of the differentiable renderer for the Dynamoc Gaussian Splat (.dgs) file format. This repository contains:

• A CUDA-accerated, differentiable renderer for dynamic gaussian splats.
• Utilties for data loading and initialization of dynamic gaussian splats.

This library serves as core infrastructure powering all of the models within Splatkit. All of Splatkit's gaussian training utilities use DDGS as their renderer.

Building + Usage

To use DDGS within your own project, you will have to build it from source. A pip install is coming soon! First, ensure you have the necessary dependencies:

• A CUDA toolkit compatible with your system's CUDA drivers
• A version of pytorch, compiled with the same CUDA version as your CUDA toolkit. See here.

Then, you can clone the repository and initialize the submodules:

git clone [email protected]:splatsdotcom/DDGS.git
cd DDGS
git submodule update --init --recursive

The project can then be built with:

pip install --no-build-isolation -e .

Now, ddgs will be available system-wide, it can be imported with a simple:

import ddgs

See example.py for an example of how to setup ddgs as well as train your first splats.

Documentation

Coming soon!