Dropping Anchor and Spherical Harmonics for Sparse-view Gaussian Splatting

2026-02-24

AI Summary

This work addresses the performance degradation of 3D Gaussian Splatting (3DGS) under sparse viewpoints, which stems from local redundancy and overfitting to high-order spherical harmonics coefficients. Existing Dropout methods are limited by neighborhood compensation effects and neglect higher-order color information. To overcome these issues, we propose an anchor-based Dropout strategy that randomly selects anchor points and simultaneously removes their spatially neighboring Gaussians, effectively suppressing compensation. Additionally, we introduce random dropping of high-order spherical harmonics coefficients to encourage appearance modeling to rely on lower-order representations. This approach is the first to incorporate an anchor mechanism into 3DGS regularization and extends Dropout to spherical harmonics-based color attributes. It significantly improves reconstruction quality under sparse views with negligible computational overhead and integrates seamlessly into various 3DGS variants.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Non-convex Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Data transparency and provenance

Abstract

Recent 3D Gaussian Splatting (3DGS) Dropout methods address overfitting under sparse-view conditions by randomly nullifying Gaussian opacities. However, we identify a neighbor compensation effect in these approaches: dropped Gaussians are often compensated by their neighbors, weakening the intended regularization. Moreover, these methods overlook the contribution of high-degree spherical harmonic coefficients (SH) to overfitting. To address these issues, we propose DropAnSH-GS, a novel anchor-based Dropout strategy. Rather than dropping Gaussians independently, our method randomly selects certain Gaussians as anchors and simultaneously removes their spatial neighbors. This effectively disrupts local redundancies near anchors and encourages the model to learn more robust, globally informed representations. Furthermore, we extend the Dropout to color attributes by randomly dropping higher-degree SH to concentrate appearance information in lower-degree SH. This strategy further mitigates overfitting and enables flexible post-training model compression via SH truncation. Experimental results demonstrate that DropAnSH-GS substantially outperforms existing Dropout methods with negligible computational overhead, and can be readily integrated into various 3DGS variants to enhance their performances. Project Website: https://sk-fun.fun/DropAnSH-GS

Problem

• 3D Gaussian Splatting
• sparse-view
• overfitting
• spherical harmonics
• Dropout

Innovation

• Gaussian Splatting
• Dropout
• Spherical Harmonics
• Sparse-view Reconstruction
• Anchor-based Regularization