AI GitHub
gsplat for Gaussian splatting research and rendering
Use differentiable Gaussian rasterization building blocks from Python.
Key Takeaways
- gsplat exposes operations used to render Gaussian scene representations and optimize them against views.
- Start with known cameras and a reference output.
- Start with a small documented scene and compare image quality, rendering time and memory across settings.
- The library is one part of scene reconstruction.
How it works
gsplat exposes operations used to render Gaussian scene representations and optimize them against views. It fits inside a training pipeline that supplies cameras, input data and objectives.
What you need
Start with known cameras and a reference output. The rendering code needs consistent shapes and coordinate conventions from the scene and camera representation.
What to try first
Start with a small documented scene and compare image quality, rendering time and memory across settings. Test camera conventions on a known input before training.
Check one controlled camera move before training a large scene. A coordinate or projection mismatch can look like a poor rasterization result.
Where to look for failures
The library is one part of scene reconstruction. Data preparation and training choices remain part of the surrounding workflow.
Related Franklin pages
Decide whether the goal is to generate an object or reconstruct a captured scene. Inspect the asset from new viewpoints and verify what your renderer can actually consume.
- Generate a 3D representation from supported image or text conditioning workflows. TRELLIS for generating 3D assets
- Train and inspect neural scene representations from captured views. Nerfstudio for reconstructing scenes from images
For video assembled from authored HTML and motion graphics, see Franklin’s HyperFrames profile. HyperFrames
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗