Gaussian Splatting Trainers Overview
This document provides a qualitative overview of Gaussian Splatting training pipelines, including both commercial and open-source tools.
The term viewer refers to the ability to interactively navigate the reconstructed scene after training, rather than to a scientific inspection tool.
Qualitative labels such as Low, Medium, and High do not represent absolute performance metrics.
They are comparative assessments derived from practical experimentation and documentation review, and should be interpreted as relative indicators within this comparison rather than objective measurements.
Overview Tables
Commercial Trainers
| Tool Type | Name | Tags | Open Source | Platform | Input | Output Format | Output Size | Processing Time | Entry Guide | Notes |
|---|---|---|---|---|---|---|---|---|---|---|
| Trainer + Viewer | Luma AI | trainer, online, commercial, web-view | No | Cloud | Video / Images | Proprietary, .ply (mesh / point cloud export) |
Low | Medium | Easy | Fully automated cloud-based pipeline |
| Trainer + Viewer | Teleport | trainer, online, commercial, immersive-view | No | Cloud | Video / Images | Proprietary | Variable | Medium | Easy | Desktop and VR-oriented cloud-based pipeline |
| Trainer + Viewer | Polycam | trainer, online, mobile, commercial, web-view | No | iOS / Android / Cloud | Video / Images | Multiple 3D formats (.ply only in paid plans) | Medium | Low | Easy | Mobile and cloud-enabled GS pipeline |
| Trainer + Viewer | Postshot | trainer, desktop, commercial, desktop-view | No | Windows | Images / Video | .ply, .spz, .psht |
High | Medium | Medium | Local desktop GS training; .ply export available only in paid plans |
| Trainer + Viewer | Scaniverse | trainer, mobile, commercial, immersive-view | No | iOS / Android | Video | .ply, .spz |
Medium | Low | Easy | On-device GS training with direct .ply export |
Open-Source Trainers
| Tool Type | Name | Tags | Open Source | Platform | Input | Output Format | Output Size | Processing Time | Entry Guide | Notes |
|---|---|---|---|---|---|---|---|---|---|---|
| Trainer | Inria gaussian-splatting | trainer, desktop, open-source | Yes | Windows / Linux | Images | .ply |
High | High | Advanced | Reference open-source implementation |
| Trainer | gsplat | trainer, backend, desktop, open-source, library | Yes | Windows / Linux | Images | .ply |
High | Medium | Advanced | High-performance PyTorch library for GS |
| Trainer + Viewer | Nerfstudio | trainer, desktop, open-source, framework | Yes | Windows / Linux | Images / Video / COLMAP | .ply |
Low | Low | Advanced | Modular framework |
| Trainer | OpenSplat | trainer, desktop, open-source | Yes | Windows / Linux | Images (COLMAP / Nerfstudio) | .ply |
Medium | Medium | Advanced | C++ native GS implementation based on LibTorch |
| Trainer + Viewer | LichtFeld Studio | trainer, desktop, open-source | Yes | Windows / Linux | Images / Video / COLMAP | .ply, .spz, .lfs |
High | Medium | Advanced | Optimized C++ trainer |
Tool Notes
Commercial Trainers
Luma AI — Fully automated cloud-based pipeline. Scenes are stored in a proprietary format optimised for real-time rendering; export to .ply is available and scenes can be explored interactively via the web interface.
Teleport (docs) — Cloud reconstruction platform by Varjo designed for spatial visualization. Allows interactive navigation of reconstructed environments in browser; .ply export available only in paid plans.
Polycam (guide) — Mobile and web capture platform supporting photogrammetry and Gaussian Splatting workflows. Models can be previewed online; point-cloud export formats (including .ply) depend on subscription tier.
Postshot (docs · video) — Local desktop GS training with configurable reconstruction parameters and integrated scene viewing after training. .ply export available only in paid plans.
Scaniverse — On-device GS training performed locally on mobile. Scenes can be interactively explored on mobile and in VR applications; direct .ply and .spz export supported.
Open-Source Trainers
Inria gaussian-splatting — Reference research implementation of the original 3D Gaussian Splatting method.
gsplat — CUDA/PyTorch library implementing Gaussian Splatting operations. Serves as backend for higher-level frameworks such as Nerfstudio.
Nerfstudio (docs) — Modular training framework for neural fields with an integrated web viewer for real-time monitoring and scene navigation. Uses gsplat as CUDA backend for its Splatfacto model.
OpenSplat — Standalone C++ GS implementation using LibTorch, removing Python dependencies for portability. Supports multiple dataset formats including COLMAP, Nerfstudio, and OpenMVG.
LichtFeld Studio (website) — Standalone C++ training suite with integrated real-time visualization and a GUI for monitoring training progress and navigating the reconstructed scene.