Counting Contributions, Not Views: Occlusion-Aware Densification for 3D Gaussian Splatting

Key points

• These statistics pass a frustum-based visibility filter blind to occlusion: fully occluded Gaussians still count as observed, diluting the averaged gradient signal and suppressing densification precisely where geometry is occluded.1
• We correct this with an occlusion-aware filter that restricts densification statistics to views in which a Gaussian actually contributes, and replace the criterion's raw view-count normalizer with an average-contribution one.1
• Integrated into a StopThePop-based framework, the two together improve every image-quality metric over the occlusion-blind baseline on occlusion-heavy scenes (up to 6% lower LPIPS), at less than half the Gaussians a view-count normalizer requires.1
• Under the longer schedules these large scenes require, the count stays bounded where most criteria diverge, and on smaller benchmarks our method stays competitive at a fraction of the competitors' count.1

Sources (1)

• [1]Counting Contributions, Not Views: Occlusion-Aware Densification for 3D Gaussian Splatting
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 06:09 PM

  3D Gaussian Splatting grows a dense scene from a sparse initialization through adaptive density control (ADC), driven by per-Gaussian statistics gathered during rendering.

  These statistics pass a frustum-based visibility filter blind to occlusion: fully occluded Gaussians still count as observed, diluting the averaged gradient signal and suppressing densification precisely where geometry is occluded.

Extractive summary: sentences quoted from the sources.