localai-org/free-splatter.cpp — explained in plain English
Analysis updated 2026-05-18
Turn a set of photos of an object into a 3D Gaussian splat scene without a GPU or Python.
Embed the free-splatter engine into another application through its plain C API.
Run a browser based demo where dropped photos are reconstructed into a viewable 3D scene.
Benchmark CPU versus GPU reconstruction speed for sparse view 3D reconstruction.
| localai-org/free-splatter.cpp | adiao1973/librobotbagfix | authrequest/freeloader | |
|---|---|---|---|
| Stars | 31 | 31 | 30 |
| Language | C++ | C++ | C++ |
| Setup difficulty | moderate | hard | hard |
| Complexity | 4/5 | 4/5 | 5/5 |
| Audience | developer | ops devops | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires downloading model weights separately from Hugging Face and a C++17 compiler.
free-splatter.cpp turns a handful of ordinary photos of an object or scene into a 3D representation called Gaussian splats, without needing camera position information, without needing a GPU, and without needing Python installed. It is a small C and C++ program that runs a neural network called FreeSplatter to do this work, and it can run on a regular computer processor or speed things up on a compatible graphics card. You give the program several photos of the same object or scene, and for every pixel in those photos it produces a small 3D shape called a Gaussian, which includes that point's position, color, transparency, size, and orientation in space. A viewer built for Gaussian splats can then render all of these points together as a 3D scene you can look around in from any angle. According to the README's own benchmarks on one test machine, using a graphics card takes about a fifth of a second to process two photos, which is roughly six times faster than the original Python version of FreeSplatter running on the same graphics card. Even running on the processor alone, with no graphics card at all, it is still about four times faster than the reference Python version. The project includes a command line tool for turning photos into splat files, plus a small web application, built with the Go programming language, that lets you drop photos in a browser and see the reconstruction, and can also stitch several reconstructed scenes together into a demo video. For developers, the core engine is also available as a plain C library, so it can be embedded directly into other software written in different programming languages. Building it requires CMake and a C++ compiler, with an optional Nix based setup that handles these dependencies automatically. Model weights are downloaded separately from Hugging Face. The project and the model weights it uses are both released under the Apache 2.0 license.
A CPU friendly C++ tool that turns a handful of photos into a 3D Gaussian splat scene, no camera poses or GPU required.
Mainly C++. The stack also includes C++, ggml, Vulkan.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.