Locality-Aware Density Control for Efficient Gaussian-based Image Representation

2026-07-20

AI Summary

Existing Gaussian-based image representation methods struggle to allocate Gaussian capacity efficiently during optimization, often resulting in fragmented under-reconstructed regions and redundant over-reconstructed areas. This work proposes a plug-and-play density control framework that, for the first time, jointly leverages the local continuity of reconstruction error in image space and the similarity among neighboring Gaussians in Gaussian space to co-optimize Gaussian allocation and redundancy removal. The approach integrates region-level Gaussian densification (RGD), similarity-driven Gaussian merging (SDGM), and a local color consistency constraint. Evaluated on benchmarks such as CLIC, it consistently enhances multiple baseline methods, achieving a 2.93 dB PSNR improvement over the GI method under a 30k Gaussian budget.

Abstract

2D Gaussian Splatting is an attractive direction for image representation due to its explicit formulation, fast rasterization, and favorable decoding efficiency. The representation quality of this paradigm depends on the proper allocation of Gaussian capacity to the demanding regions. However, existing methods fail to allocate Gaussian capacity efficiently during optimization: under-reconstructed content is often refined in a fragmented pixel-wise manner, while neighboring optimized Gaussians with similar attributes are redundantly retained. This inefficiency motivates the need for a density control framework that jointly addresses insufficient allocation in under-reconstructed regions and redundant allocation in over-reconstructed regions. Our key insight is that this framework should exploit two complementary forms of locality: the local continuity of reconstruction errors in image space for improved Gaussian allocation, and the local similarity of neighboring Gaussians in Gaussian space for redundant elimination. Based on this insight, we propose Locality-Aware Density Control (LocoADC), a plug-and-play framework that improves Gaussian capacity utilization through Region-wise Gaussian Densification (RGD) and Similarity-Driven Gaussian Merging (SDGM) strategies, together with a local color consistency constraint for more reliable merging. Extensive experiments on diverse datasets show that LocoADC consistently improves multiple baselines by enabling more effective local Gaussian allocation, including a 2.93 dB PSNR gain over GI on the CLIC dataset under the same 30k Gaussian budget. Code is available at: \textit{https://github.com/ChenJiaCong-1005/LocoADC}.

Problem

• Gaussian-based image representation
• density control
• locality-aware
• redundancy elimination
• capacity allocation

Innovation

• Locality-Aware Density Control
• Gaussian Splatting
• Region-wise Densification
• Similarity-Driven Merging
• Image Representation