Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Gaussian What ⁉️


Rendering Gaussian Splats on the Web

Fivos
DOGANIS

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Seminal Paper

3D Gaussian Splatting for Real-Time Radiance Field Rendering

INRIA, SIGGRAPH 2023

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Today

Christoph SCHINDELAR : QUECHUA - Webviewer
Tipatat CHENNAVASIN : flowers
Adam HAVKIN : Lowy Concert Hall
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

3D Points

cover picture by Luma AI, taken from here
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Sparse Point Cloud produced by SfM

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

3D Points ➡️ Gaussian Splats

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Gaussian Splats

  • 3D points / particles ➡️ blobs 3D Gaussians projected in 2D

  • Rendering Steps

    • Project each gaussian into 2D from the camera perspective.
    • Sort the gaussians by depth ⏳
    • For each pixel
      • iterate over each gaussian front-to-back
      • blend them together
  • See training and rasterization details

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
fit:50%
fit:50%
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Gaussian Splats ⭐

  • 3D points / particles ➡️ blobs 3D Gaussians projected in 2D
  • each particle has a
    • position: x, y, z (mean μ\mu)
    • rotation + non-uniform scale: Mat3 (covariance Σ\Sigma)
    • opacity (sigmoid σ(α)\sigma(\alpha))
    • view-dependent color: r, g, b + SH Coeffs (Spherical Harmonics)
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Spherical Harmonics

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Spherical Harmonics (SH)

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

File Formats

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

.PLY file format

  • similar to RGB point cloud
    • basic color described using f_dc_ values
  • optional coefficients and parameters
  • nx, ny, nz unused
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
ply
format binary_little_endian 1.0
element vertex 1534456
property float x
property float y
property float z
property float nx
property float ny
property float nz
property float f_dc_0
property float f_dc_1
property float f_dc_2
property float f_rest_0
(... f_rest from 1 to  43...)
property float f_rest_44
property float opacity
property float scale_0
property float scale_1
property float scale_2
property float rot_0
property float rot_1
property float rot_2
property float rot_3
end_header
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Other formats

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Drawbacks

  • file size (~100 MB per scene, if unoptimized)
  • rendering order is important
    • sorting primitives ➡️ performance issues
    • popping artifacts
  • optimizing can take a long time
  • models need to be cleaned up
    • "floaters"
  • no real geometry (for collision detection, relighting etc)
    • but convex hull mesh can be created from point cloud
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Web Viewers

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Web Libraries

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Main Web implementations

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
<!DOCTYPE html>
<html>
  <head>
    <script src="https://aframe.io/releases/1.4.2/aframe.min.js"></script>
    <script src="https://quadjr.github.io/aframe-gaussian-splatting/index.js"></script>
  </head>
  <body>
    <a-scene renderer="antialias: false">
      <a-entity rotation="10 0 0">
        <a-entity position="1.2 1.2 -2.7" 
        animation="property: rotation; to: 0 360 0; dur: 10000; easing: linear; loop: true">
          <a-sphere position="0 0 0.5" radius="0.2" color="#EF2D5E"></a-sphere>
          <a-box position="0.5 0 0" rotation="0 45 0" height="0.4" width="0.4" 
          depth="0.4" color="#4CC3D9" shadow></a-box>
          <a-cylinder position="0 0 -0.5" radius="0.25" height="0.4" color="#FFC65D" shadow></a-cylinder>
        </a-entity>
      </a-entity>
      <a-entity gaussian_splatting="src: https://huggingface.co/cakewalk/splat-data/resolve/main/truck.splat;" 
      rotation="0 0 0" position="0 1.5 -2"></a-entity>
      <a-sky color="#000"></a-sky>
    </a-scene>
  </body>
</html>

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Gaussian Splats + Standard WebGL rendering

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Other Tools

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Postshot by Jawset

  • Free Desktop app (beta)
Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

The End

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

Questions?

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025

The Future

Fivos Doganis | Presentation licensed under CC BY-SA 4.0 | 2025
Ahmad EMRANIAN : Akmal Face

https://marpit.marp.app/directives?id=tweak-theme-style

https://www.connectedpapers.com/main/2cc1d857e86d5152ba7fe6a8355c2a0150cc280a/graph

![bg 70%](https://miro.medium.com/v2/resize:fit:1400/format:webp/0*QmfJf4pJH9zfenfA.png)

![bg 60%](https://towardsdatascience.com/wp-content/uploads/2023/12/0EGVALNS9pxPyXQ8N.png)