GaussianImage++: Boosted Image Representation and Compression with 2D Gaussian Splatting

2025-12-22

AI Summary

To address the redundancy of Gaussian primitives and the trade-off between fidelity and compression in 2D Gaussian splatting (GS) image representations, this paper proposes an efficient implicit image representation. Our method introduces: (1) a distortion-driven progressive density growth mechanism that dynamically spawns Gaussians according to signal importance; (2) content-adaptive Gaussian filters to enhance local structural reconstruction; and (3) an attribute-separate learnable scalar quantizer with quantization-aware training for compact encoding and end-to-end optimization. Experiments demonstrate that our approach achieves superior PSNR and SSIM over GaussianImage and INR-based COIN using fewer than 1,000 Gaussians. It enables real-time decoding and low memory footprint, striking a significant balance between image compression efficiency and implicit representational fidelity.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Representation Learning
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Implicit neural representations (INRs) have achieved remarkable success in image representation and compression, but they require substantial training time and memory. Meanwhile, recent 2D Gaussian Splatting (GS) methods ( extit{e.g.}, GaussianImage) offer promising alternatives through efficient primitive-based rendering. However, these methods require excessive Gaussian primitives to maintain high visual fidelity. To exploit the potential of GS-based approaches, we present GaussianImage++, which utilizes limited Gaussian primitives to achieve impressive representation and compression performance. Firstly, we introduce a distortion-driven densification mechanism. It progressively allocates Gaussian primitives according to signal intensity. Secondly, we employ context-aware Gaussian filters for each primitive, which assist in the densification to optimize Gaussian primitives based on varying image content. Thirdly, we integrate attribute-separated learnable scalar quantizers and quantization-aware training, enabling efficient compression of primitive attributes. Experimental results demonstrate the effectiveness of our method. In particular, GaussianImage++ outperforms GaussianImage and INRs-based COIN in representation and compression performance while maintaining real-time decoding and low memory usage.

Problem

• Reduces Gaussian primitives for high-fidelity image representation
• Optimizes primitives with context-aware filters for varied content
• Compresses attributes efficiently using quantization-aware training

Innovation

• Distortion-driven densification mechanism for Gaussian primitives
• Context-aware Gaussian filters optimize primitive allocation
• Attribute-separated learnable scalar quantizers enable efficient compression