Decomposing Densification in Gaussian Splatting for Faster 3D Scene Reconstruction

2025-07-27

AI Summary

To address slow convergence in 3D Gaussian Splatting (GS) training caused by inefficient Gaussian primitive densification and suboptimal spatial distribution, this paper proposes an energy-guided global-to-local densification strategy integrated with a multi-resolution training framework. By decoupling splitting and cloning operations—revealing their distinct roles in detail fidelity versus computational efficiency—we design a staged, energy-driven Gaussian growth mechanism that significantly improves distribution rationality. Our method jointly leverages energy density analysis, dynamic pruning, and coarse-to-fine optimization. Evaluated on multiple standard benchmarks, it achieves over 2× training acceleration while using fewer Gaussians to attain superior reconstruction quality—yielding PSNR/SSIM gains of 1.2–2.5 dB. This work establishes a new paradigm for efficient, high-fidelity 3D scene reconstruction.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments

Abstract

3D Gaussian Splatting (GS) has emerged as a powerful representation for high-quality scene reconstruction, offering compelling rendering quality. However, the training process of GS often suffers from slow convergence due to inefficient densification and suboptimal spatial distribution of Gaussian primitives. In this work, we present a comprehensive analysis of the split and clone operations during the densification phase, revealing their distinct roles in balancing detail preservation and computational efficiency. Building upon this analysis, we propose a global-to-local densification strategy, which facilitates more efficient growth of Gaussians across the scene space, promoting both global coverage and local refinement. To cooperate with the proposed densification strategy and promote sufficient diffusion of Gaussian primitives in space, we introduce an energy-guided coarse-to-fine multi-resolution training framework, which gradually increases resolution based on energy density in 2D images. Additionally, we dynamically prune unnecessary Gaussian primitives to speed up the training. Extensive experiments on MipNeRF-360, Deep Blending, and Tanks & Temples datasets demonstrate that our approach significantly accelerates training,achieving over 2x speedup with fewer Gaussian primitives and superior reconstruction performance.

Problem

• Slow convergence in Gaussian Splatting training
• Inefficient densification of Gaussian primitives
• Suboptimal spatial distribution of 3D elements

Innovation

• Global-to-local densification strategy for Gaussians
• Energy-guided coarse-to-fine multi-resolution training
• Dynamic pruning of unnecessary Gaussian primitives