Occlusion aware method improves 3d scene densification in Gaussian splatting
Counting Contributions, Not Views: Occlusion-Aware Densification for 3D Gaussian Splatting
Computer Vision and Pattern RecognitionGraphics
Summary
Creating detailed 3D scenes from sparse starting points can be challenging when parts of the scene are hidden by others, called occlusion. The authors found that previous methods mistakenly treated hidden parts as visible, which slowed down improving detail in those areas. They developed a way to check if a part of the scene really contributes to the visible view before updating it. This approach leads to clearer 3D images with fewer parts needed, especially in scenes with many hidden objects.
What this means in practice
- •For 3d graphics developers: Improve rendering quality and efficiency for complex 3D scenes with many occluded objects by using occlusion-aware densification.
- •For virtual reality content creators: Create more detailed and visually accurate 3D environments by better managing scene elements hidden behind others.
Authors
Marcel Rogge, René Schuster, Didier Stricker
Abstract
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. 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. 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. 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.