Training Tool

Brush

Brush is an open-source Gaussian Splatting project for real-time training and inspection, useful for people who want a more interactive route than classic research scripts.

Stage: TrainingInteractive trainingOpen-source workflowsFast inspection

Open official linkGitHub Repository->

What It Does

Brush is interesting because it pushes Gaussian Splatting training toward a more immediate, visual workflow. Instead of treating training as a black box that only produces an output folder, it encourages inspection while the scene is taking shape.

It is a good page to keep on your radar if you prefer open-source tools but want a friendlier loop than launching long training scripts and checking results afterward.

How To Use It In 3DGS

Prepare a clean aligned dataset before opening the training workflow.
Start with a small capture so you can learn the controls and quality tradeoffs quickly.
Inspect floaters, coverage holes, and unstable areas while the model develops.
Export to a viewer-friendly format and compare with other trainers on the same dataset.

Things To Watch

Project capabilities may move quickly, so check the repository before building a long-term pipeline around a specific command.
Interactive training still depends on good camera poses and enough GPU headroom.
If you need scripted reproducibility, document versions and settings carefully.

Official And Related Links

GitHub Repository 3D Gaussian Splatting Paper COLMAP Documentation

What Makes Brush Different

Brush targets macOS, Windows, Linux, AMD/NVIDIA/Intel GPUs, Android, and browser use by building around WebGPU-compatible technology and the Burn machine learning framework.
Its Rust/WGSL codebase makes it interesting for cross-platform experiments where CUDA-only trainers are too restrictive.
The repository notes rendering and training benchmarks against gsplat kernels, so it is worth watching if you care about portable training performance.