brush
SHA-256A cross-platform 3D reconstruction engine using Gaussian splatting, supporting macOS, Windows, Linux, Android, and browser. Unlike gsplat, it requires no CUDA dependencies and runs on all GPUs.
Smart Download
Download Download Version
v0.3.0 · 151.4 MB
A CUDA-free 3D reconstruction engine using Gaussian splatting, runs on all platforms and browsers.
Core Features
- Cross-platform native support: macOS, Windows, Linux, Android, and web (Chrome/Edge)
- No CUDA required – uses WebGPU and Burn, works on AMD/Nvidia/Intel
- Real-time interactive training with live view comparison
- Supports COLMAP and Nerfstudio formats, masks, and animation sequences
- Built-in high-performance viewer for .ply/.compressed.ply and URL streaming
What It Can't Do
- •Web demo currently works only on Chrome and Edge; Firefox and Safari support pending
- •Training requires at least 4GB VRAM for decent performance
- •COLMAP data should have moderate sparsity; extremely large point clouds may impact interactivity
Use Cases
- Real-time 3D reconstruction and visualization for film, gaming, and architecture
- Showcasing Gaussian splatting models on mobile or web without a dedicated GPU
- Research and education – quickly iterate on new datasets and algorithms
Detailed Introduction
Brush is a real-time 3D reconstruction engine built on Gaussian splatting and the Burn machine learning framework. It works natively on macOS, Windows, Linux, Android, and in the browser via WebGPU, eliminating the need for CUDA libraries. Compared to gsplat, Brush offers simpler deployment without dependency-heavy packages, supports AMD/Nvidia/Intel GPUs equally, and provides an interactive viewer that can train and render scenes in real time. It ingests COLMAP or Nerfstudio data, supports masking, animation playback of splat sequences, and offers a CLI for headless usage. Benchmarks show rendering and training performance typically faster than gsplat.
Troubleshooting & FAQ (1)
feature inquiryHow to train on fisheye images (e.g., Zipnerf) in Brush?
Brush now supports fisheye training, including OPENCV_FISHEYE, thanks to the implementation in PR #434. You can train directly on datasets with fisheye distortion and use masks.
Tags
Getting Started
Install the software
Double-click the downloaded installer and follow the prompts
Install Rust 1.88+ and platform build tools
Run cargo run --release from the project root to build an optimized binary
Launch the app and load COLMAP data or .ply files via CLI or UI to train/view
- Install Rust 1.88+ and platform build tools
- Run cargo run --release from the project root to build an optimized binary
- Launch the app and load COLMAP data or .ply files via CLI or UI to train/view
SHA-256 checksum verified
Checksum extracted from GitHub official Release page
SHA256 Checksum
b68e3e9cf052d51bf3ee30776fa5a364de7f2ba13b58443128ff797bb7bcfcd6This checksum is extracted from the GitHub Release page. Verify file integrity after download.
All SHA-256 checksums on this platform are extracted from the project's official GitHub Release page, without any modification. You can independently verify them on the GitHub Releases page.
Open Source Transparency
View GitHub SourceUninstall Info
Simply delete the Brush executable and project folder. If installed via Cargo, run cargo uninstall brush.
No Extra Dependencies
Ready to use after download. No additional runtime required.
Having issues? Check the FAQ below
1 FAQ
Similar Projects
ollama
Ollama lets you download, run, and manage large language models locally. One command, multiple platforms, endless possibilities.
llama.cpp
High-performance LLM inference engine in C/C++ with minimal dependencies, supporting quantized models (1.5–8 bit) and diverse hardware (Apple Silicon, CUDA, Vulkan, etc.).
opencv
OpenCV is an open-source computer vision and machine learning library with over 2500 optimized algorithms for real-time image and video analysis.