# graphdeco-inria/gaussian-splatting architecture

> Architecture diagram of the GitHub repository graphdeco-inria/gaussian-splatting, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-09-23). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/graphdeco-inria/gaussian-splatting
- Repository: https://github.com/graphdeco-inria/gaussian-splatting

## Overview

This repository implements 3D Gaussian Splatting: prepare or supply multi-view images and SfM data, optimize a Gaussian scene representation, then render novel views and evaluate image quality. Training also supports depth regularization and a network-connected interactive viewer; the README describes a separate SIBR real-time viewer. The sampled source confirms the training, scene-loading, rasterization, and GUI paths. Evaluation scripts and SIBR internals were not sampled, so their detailed wiring is not inferred.

Main entry points: [`train.py`](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/train.py) (Training loop), [`gaussian_renderer/__init__.py`](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/gaussian_renderer/__init__.py) (Gaussian renderer), [`convert.py`](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/convert.py) (SfM converter), [`metrics.py`](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/metrics.py) (Image metrics)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_prep["Dataset preparation"]
  node_converter["SfM converter<br/>[convert.py]"]
  node_sceneio["Scene loading<br/>[dataset_readers.py]"]
  node_colmapio["COLMAP model I/O"]
  node_depthscale["Depth calibration"]
  node_scene["Scene assembly<br/>[__init__.py]"]
end

subgraph group_train["Optimization"]
  node_trainloop["Training loop<br/>[train.py]"]
  node_params["Training parameters<br/>[__init__.py]"]
  node_cameras["Camera model<br/>[cameras.py]"]
  node_gaussians["Gaussian model<br/>[gaussian_model.py]"]
end

subgraph group_render["Rendering and viewing"]
  node_renderer["Gaussian renderer<br/>[__init__.py]"]
  node_rasterizer["CUDA rasterizer"]
  node_netgui["Training network GUI<br/>[network_gui.py]"]
  node_rendercli["Novel-view rendering<br/>[render.py]"]
  node_sibr["SIBR real-time viewer"]
end

subgraph group_eval["Evaluation"]
  node_full_eval["Evaluation runner<br/>[full_eval.py]"]
  node_metrics["Image metrics<br/>[metrics.py]"]
  node_lpips["LPIPS metric<br/>[__init__.py]"]
end

node_user(("Researcher"))
node_images["Input images"]
node_colmap["COLMAP"]
node_model["Trained Gaussian model"]
node_views["Rendered views"]
node_reference["Reference images"]

node_user -->|"prepares dataset"| node_converter
node_images -->|"supplies images"| node_converter
node_converter -->|"runs SfM"| node_colmap
node_colmap -->|"returns reconstruction"| node_converter
node_converter -->|"writes dataset"| node_sceneio
node_sceneio -->|"reads model"| node_colmapio
node_sceneio -->|"builds scene data"| node_scene
node_user -->|"starts training"| node_trainloop
node_trainloop -->|"loads settings"| node_params
node_trainloop -->|"loads scene"| node_scene
node_trainloop -->|"optimizes"| node_gaussians
node_trainloop -->|"renders training views"| node_renderer
node_renderer -->|"rasterizes Gaussians"| node_rasterizer
node_renderer -->|"reads parameters"| node_gaussians
node_renderer -->|"uses camera transforms"| node_cameras
node_trainloop -->|"serves viewer requests"| node_netgui
node_netgui -->|"requests custom render"| node_renderer
node_trainloop -->|"saves checkpoints"| node_model
node_user -->|"requests views"| node_rendercli
node_rendercli -->|"renders camera views"| node_renderer
node_rendercli -->|"loads model"| node_model
node_renderer -->|"produces images"| node_views
node_user -->|"views scene"| node_sibr
node_model -.->|"loads for viewing"| node_sibr
node_user -->|"runs evaluation"| node_full_eval
node_full_eval -->|"requests renders"| node_rendercli
node_full_eval -->|"computes scores"| node_metrics
node_metrics -->|"computes perceptual distance"| node_lpips
node_metrics -->|"compares renders"| node_views
node_metrics -->|"compares references"| node_reference
node_user -->|"calibrates depth"| node_depthscale
node_depthscale -->|"reads SfM model"| node_colmapio
node_depthscale -->|"writes depth parameters"| node_sceneio

click node_converter "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/convert.py"
click node_sceneio "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/dataset_readers.py"
click node_colmapio "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/utils/read_write_model.py"
click node_depthscale "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/utils/make_depth_scale.py"
click node_scene "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/__init__.py"
click node_trainloop "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/train.py"
click node_params "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/arguments/__init__.py"
click node_cameras "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/cameras.py"
click node_gaussians "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/gaussian_model.py"
click node_renderer "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/gaussian_renderer/__init__.py"
click node_rasterizer "https://github.com/graphdeco-inria/gaussian-splatting/tree/main/submodules/diff-gaussian-rasterization"
click node_netgui "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/gaussian_renderer/network_gui.py"
click node_rendercli "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/render.py"
click node_sibr "https://github.com/graphdeco-inria/gaussian-splatting/tree/main/SIBR_viewers"
click node_full_eval "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/full_eval.py"
click node_metrics "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/metrics.py"
click node_lpips "https://github.com/graphdeco-inria/gaussian-splatting/blob/main/lpipsPyTorch/__init__.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_converter,node_sceneio,node_colmapio,node_depthscale,node_scene,node_views toneBlue
class node_trainloop,node_params,node_cameras,node_gaussians toneAmber
class node_renderer,node_rasterizer,node_netgui,node_rendercli,node_sibr toneMint
class node_full_eval,node_metrics,node_lpips toneRose
class node_user,node_images,node_colmap,node_model,node_reference toneIndigo
```

## Components

### Dataset preparation

- [SfM converter](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/convert.py): `convert.py`
- [Scene loading](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/dataset_readers.py): `scene/dataset_readers.py`
- [COLMAP model I/O](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/utils/read_write_model.py): `utils/read_write_model.py`
- [Depth calibration](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/utils/make_depth_scale.py): `utils/make_depth_scale.py`
- [Scene assembly](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/__init__.py): `scene/__init__.py`

### Optimization

- [Training loop](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/train.py): `train.py`
- [Training parameters](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/arguments/__init__.py): `arguments/__init__.py`
- [Camera model](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/cameras.py): `scene/cameras.py`
- [Gaussian model](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/scene/gaussian_model.py): `scene/gaussian_model.py`

### Rendering and viewing

- [Gaussian renderer](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/gaussian_renderer/__init__.py): `gaussian_renderer/__init__.py`
- [CUDA rasterizer](https://github.com/graphdeco-inria/gaussian-splatting/tree/main/submodules/diff-gaussian-rasterization): `submodules/diff-gaussian-rasterization`
- [Training network GUI](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/gaussian_renderer/network_gui.py): `gaussian_renderer/network_gui.py`
- [Novel-view rendering](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/render.py): `render.py`
- [SIBR real-time viewer](https://github.com/graphdeco-inria/gaussian-splatting/tree/main/SIBR_viewers): `SIBR_viewers`

### Evaluation

- [Evaluation runner](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/full_eval.py): `full_eval.py`
- [Image metrics](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/metrics.py): `metrics.py`
- [LPIPS metric](https://github.com/graphdeco-inria/gaussian-splatting/blob/main/lpipsPyTorch/__init__.py): `lpipsPyTorch/__init__.py`

### Other components

- **Researcher**
- **Input images**
- **COLMAP**
- **Trained Gaussian model**
- **Rendered views**
- **Reference images**

## Connections

- Researcher → SfM converter: prepares dataset
- Input images → SfM converter: supplies images
- SfM converter → COLMAP: runs SfM
- COLMAP → SfM converter: returns reconstruction
- SfM converter → Scene loading: writes dataset
- Scene loading → COLMAP model I/O: reads model
- Scene loading → Scene assembly: builds scene data
- Researcher → Training loop: starts training
- Training loop → Training parameters: loads settings
- Training loop → Scene assembly: loads scene
- Training loop → Gaussian model: optimizes
- Training loop → Gaussian renderer: renders training views
- Gaussian renderer → CUDA rasterizer: rasterizes Gaussians
- Gaussian renderer → Gaussian model: reads parameters
- Gaussian renderer → Camera model: uses camera transforms
- Training loop → Training network GUI: serves viewer requests
- Training network GUI → Gaussian renderer: requests custom render
- Training loop → Trained Gaussian model: saves checkpoints
- Researcher → Novel-view rendering: requests views
- Novel-view rendering → Gaussian renderer: renders camera views
- Novel-view rendering → Trained Gaussian model: loads model
- Gaussian renderer → Rendered views: produces images
- Researcher → SIBR real-time viewer: views scene
- Trained Gaussian model → SIBR real-time viewer: loads for viewing
- Researcher → Evaluation runner: runs evaluation
- Evaluation runner → Novel-view rendering: requests renders
- Evaluation runner → Image metrics: computes scores
- Image metrics → LPIPS metric: computes perceptual distance
- Image metrics → Rendered views: compares renders
- Image metrics → Reference images: compares references
- Researcher → Depth calibration: calibrates depth
- Depth calibration → COLMAP model I/O: reads SfM model
- Depth calibration → Scene loading: writes depth parameters
