Portable Trainer

OpenSplat

OpenSplat is a free and open-source C++ implementation of 3D Gaussian Splatting focused on portability across Windows, macOS, and Linux.

Stage: TrainingCross-platform useC++ pipelinesCPU fallback

Open official linkGitHub Repository->

What It Does

OpenSplat is a practical option when you need a native trainer outside a Python-heavy stack. Its broad input support also makes it useful for photogrammetry and mapping pipelines that do not start in COLMAP.

Because it can run on CPU, it has a wider hardware story than many CUDA-only projects, though GPU training is still the realistic choice for serious scenes.

How To Use It In 3DGS

Create or collect a sparse reconstruction from COLMAP, OpenSfM, ODM, OpenMVG, or Nerfstudio.
Run OpenSplat on the aligned data and choose an output format suited to your viewer.
Use GPU acceleration whenever possible for reasonable training time.
Open the resulting PLY or SPLAT in a viewer or editor for inspection and delivery conversion.

Things To Watch

CPU training is useful as a fallback, not as the happy path for large datasets.
Input format support is broad, but you still need valid camera poses and sparse points.
Check licensing and distribution expectations if you embed it in a commercial product.

Official And Related Links

GitHub Repository COLMAP Documentation OpenDroneMap

Why Choose It

OpenSplat is attractive when the capture source comes from mapping tools such as ODM, OpenSfM, or OpenMVG rather than a pure COLMAP folder.
Apple Metal and AMD support make it a more practical option for mixed hardware teams than CUDA-only pipelines.
The output is intentionally simple: produce PLY or SPLAT, then hand off to editors, converters, or viewers.