Gaussian splatting

Version 5 of 5

Fitted image.

CPU · 2D Gaussians · 2026

This version

An image built from 1,400 stretched 2D Gaussians, added coarse to fine like an optimiser at work: big soft ones for the background, thin ones along the edges.

How this version works

At 240 pixels wide, this version builds a static image from 1,400 seeded, procedurally placed 2D Gaussians coloured from a three-centre blob field. It allocates 260 splats to the background, 520 to the interior and 620 near the boundary, where thin ellipses align with the field's contour. Splats are sorted by their major-axis size and alpha-blended from coarse to fine into three persistent floating-point buffers. A ten-second cycle reveals them over six seconds, holds the completed image, then clears and starts again.

The trick in this version

Because the splats never move and their compositing order is fixed, each frame only blends newly revealed splats. The apparent fitting process is a predetermined sampling scheme, not a live optimiser, and Gaussian evaluation is restricted to bounded pixel regions.

Historical context · Emerging

  1. 2026 · With Love · I see you

    Gaussians in a 4K

    A 4K executable still whose author confirms about 450 fitted Gaussian splats, fitting an image, not a 3D fly-through. The scene is only starting to use them.

  2. 2024 · Equinor summer interns

    Interactive splat scenes

    Gaussian objects mixed with ordinary 3D objects, including a bowling demo with splat pins, interaction beyond passive viewing.

  3. 2023 · Kerbl, Kopanas, Leimkühler & Drettakis · SIGGRAPH 2023

    3D Gaussian Splatting

    Anisotropic Gaussians optimised against photos render captured scenes at 100+ fps at 1080p. The original implementation uses CUDA, not the fixed rasterisation pipeline.

Platform
CPU · 2D Gaussians
Era
Emerging
Rendered on
CPU
Tags
2D · coarse to fine
Sources
Gaussians in a 4K (2026)Interactive splat scenes (2024)3D Gaussian Splatting (2023)