Structure-Guided Allocation of 2D Gaussians for Image Representation and Compression

TL;DR

This work targets the bottleneck of rate-distortion efficiency in 2D Gaussian Splatting (2DGS) by introducing a structure-guided allocation principle that ties representation capacity and quantization precision to image structure. It implements three core components—structure-guided initialization (SGI), adaptive bitwidth quantization (ABQ), and geometry-consistent regularization (GCR)—to concentrate Gaussians and precision in perceptually important regions while preserving fast, resolution-independent decoding. The approach yields substantial RD gains over GSImage, with BD-rate reductions of about 43.44% on Kodak and 29.91% on DIV2K$\times2$, and decoding speeds exceeding 1{,}000 FPS, illustrating practical viability for real-time applications. Overall, the method advances 2DGS by integrating structural cues into both capacity planning and quantization, achieving superior reconstruction fidelity and compression efficiency without sacrificing speed.

Abstract

Recent advances in 2D Gaussian Splatting (2DGS) have demonstrated its potential as a compact image representation with millisecond-level decoding. However, existing 2DGS-based pipelines allocate representation capacity and parameter precision largely oblivious to image structure, limiting their rate-distortion (RD) efficiency at low bitrates. To address this, we propose a structure-guided allocation principle for 2DGS, which explicitly couples image structure with both representation capacity and quantization precision, while preserving native decoding speed. First, we introduce a structure-guided initialization that assigns 2D Gaussians according to spatial structural priors inherent in natural images, yielding a localized and semantically meaningful distribution. Second, during quantization-aware fine-tuning, we propose adaptive bitwidth quantization of covariance parameters, which grants higher precision to small-scale Gaussians in complex regions and lower precision elsewhere, enabling RD-aware optimization, thereby reducing redundancy without degrading edge quality. Third, we impose a geometry-consistent regularization that aligns Gaussian orientations with local gradient directions to better preserve structural details. Extensive experiments demonstrate that our approach substantially improves both the representational power and the RD performance of 2DGS while maintaining over 1000 FPS decoding. Compared with the baseline GSImage, we reduce BD-rate by 43.44% on Kodak and 29.91% on DIV2K.

Figures (12)

• Figure 1: BD-rate saving vs. Decoding Speed on Kodak dataset. Our method achieves the best BD-rate at low bitrates with similar decoding speed to GSImage. The upper-right region indicates better performance.
• Figure 2: Representation pipeline of our proposed method. We begin with structure-guided initialization, which initialize 2D Gaussians based on local structural complexity derived from the segmentation map and gradient map. These Gaussians are rasterized to reconstruct the image, and their attributes are optimized using a joint loss combining MSE loss $\mathcal{L}_2$ with geometry-consistent regularization $\mathcal{L}_g$.
• Figure 3: Visualization of initial 2D Gaussians‘ distribution. Our method adaptively allocates more Gaussians to complex regions. The first row shows kodim03 (3k Gaussians) and the second row shows kodim20 (7k Gaussians)
• Figure 4: Performance comparison of various initialization methods on the Kodak dataset, with SGI denoting our proposed Structure-Guided Initialization.
• Figure 5: Compression pipeline of the proposed method. After image overfitting, quantization-aware fine-tuning is performed to remove parameter redundancy and enable efficient compression. We introduce adaptive bitwidth quantization for covariance parameters, allocating higher precision to small-scale Gaussians in complex regions while using lower precision elsewhere. In addition, learned scale quantization is applied to position vector, and residual vector quantization is employed for color coefficients.