Florian-Barthel/splatviz — Gitpedia

<div align="center"> <h1> <img src="resources/images/icon.png" width="32" alt="splatviz icon"> splatviz </h1> <p> <strong>Interactive 3D Gaussian Splatting viewer and editor.</strong> </p> <p> <img alt="GitHub top language" src="https://img.shields.io/github/languages/top/Florian-barthel/splatviz"> <img alt="GitHub release" src="https://img.shields.io/github/v/release/Florian-Barthel/splatviz"> <img alt="GitHub last commit" src="https://img.shields.io/github/last-commit/Florian-Barthel/splatviz"> <img alt="Linux" src="https://img.shields.io/badge/Platform-Linux-green"> <img alt="Windows" src="https://img.shields.io/badge/Platform-Windows-green"> </p> <p> Edit splats live, compare scenes side by side, attach to training runs, and inspect render data with Python snippets. </p> <p> <a href="https://www.youtube.com/watch?v=AJGSa_wnZS8&ab_channel=FlorianBarthel">Demo video</a> · <a href="https://drive.google.com/drive/folders/1RNMis5ibSrl2a87ZX614ZPsHKaY6ams6?usp=sharing">Example scenes</a> · <a href="https://github.com/Florian-Barthel/ganviz">Check out ganviz</a> </p> <img src="resources/images/teaser.gif" width="720" alt="splatviz teaser"> </div>

Install

Install uv with curl -LsSf https://astral.sh/uv/install.sh | sh and a matching CUDA Toolkit:
11.8,
12.6,
12.8, or
13.0.

bash
git clone https://github.com/Florian-Barthel/splatviz.git
cd splatviz

Sync the environment with the CUDA extra for your system:

bash
uv sync --extra cu118
uv sync --extra cu126
uv sync --extra cu128
uv sync --extra cu130

Then install the rasterizer:

bash
uv pip install --no-build-isolation git+https://github.com/ashawkey/diff-gaussian-rasterization.git

On Windows, you may also need the 2022 Visual Studio
C++ build tools.

winget install Microsoft.VisualStudio.2022.BuildTools

Launch

bash
python run_main.py

If OpenGL picks the wrong version, set:

bash
export MESA_GL_VERSION_OVERRIDE=3.3

Attach to a running 3DGS training process:

bash
python run_main.py --mode=attach

By default, splatviz listens on 127.0.0.1:6007. Override with --host and
--port. For live editing during training, see
splatviz_network.

Widgets

<table> <tr> <td width="50%"> <h3>Edit</h3> <p>Run Python just before rendering to modify the Gaussian object live.</p> <img src="resources/images/edit.png" width="360" alt="Edit widget"> </td> <td width="50%"> <h3>Evaluate</h3> <p>Inspect render variables and visualize values with histograms.</p> <img src="resources/images/eval.png" width="360" alt="Evaluate widget"> </td> </tr> </table>

Recent Features

Version 1.3.0

2026-05: Video widget
2026-05: Easier install with uv
2026-05: External edit text support
2026-05: Resizable render window

Version 1.2.0

2025-09: Fixed eval widget and rasterizer import
2025-09: Simpler camera and file-browser scene loading
2025-09: Moved GAN mode to ganviz
2025-05: Added GAN mode

Version 1.1.0

2024-08: Training widget and live 3DGS attach mode
2024-08: Rendering refactor and improved mouse controls
2024-07: Slider presets, performance widget, and editor tooltips
2024-07: Multi-scene comparison

Citation

If splatviz helps your work, please cite:

bibtex
@misc{barthel2024gaussian,
    title={Gaussian Splatting Decoder for 3D-aware Generative Adversarial Networks},
    author={Florian Barthel and Arian Beckmann and Wieland Morgenstern and Anna Hilsmann and Peter Eisert},
    year={2024},
    eprint={2404.10625},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}