Daily papers

Monday, 9 November 2026 · 7 min read

Gaussian Spacing Normalization improves feed-forward 3D Gaussian Splatting generalization

SplatAnything introduces Gaussian Spacing Normalization (GSNorm) to canonicalize Gaussian spacing across scenes, resolving supervision ambiguity in feed-forward 3D Gaussian Splatting and achieving state-of-the-art results on a new cross-dataset benchmark.

Based on the paper SplatAnything: Gaussian Spacing Normalization for Feed-forward 3DGS across Arbitrary Scenes by Hao Yu, Zhenwei Wang, Yongzhen Hu, Yuning Gong and 6 others

Challenges in feed-forward 3D Gaussian Splatting across diverse scenes and views

Feed-forward 3D Gaussian Splatting (3DGS) methods predict a 3D Gaussian representation directly from input images without costly per-scene optimization, enabling efficient large-scale 3D reconstruction. However, achieving a single model that generalizes well across arbitrary scene scales, camera configurations, and novel viewpoints remains difficult.

A common approach uses a geometry foundation model to predict per-pixel depth and then predicts Gaussian attributes in a radial-distance-normalized coordinate space. This normalization rescales the scene based on the mean radial distance of 3D points to the camera origin, but it does not normalize the spacing between neighboring Gaussian centers, called Gaussian spacing.

Because Gaussian scale should match local Gaussian spacing for adequate surface coverage, variations in Gaussian spacing across scenes and camera setups cause shifts in Gaussian-scale targets. This leads to ambiguous supervision, where visually similar regions correspond to different Gaussian scales, causing under- or over-coverage and limiting generalization across scenes and viewpoints.

Sign in to read the whole paper

You have read the opening. The rest is free with a Bezen account, and no card is needed.

In the rest of this paper

  • How the method works, step by step, with a flow chart
  • The key formula from the paper, explained symbol by symbol
  • Three figures from the paper, and how to read them
  • How it was tested, what was found, and where it falls short
Gaussian Spacing Normalization improves feed-forward 3D Gaussian Splatting generalization · Bezen