P-4DGS: Predictive 4D Gaussian Splatting with 90$ imes$ Compression

2025-10-11

AI Summary

Existing dynamic 3D Gaussian Splatting (4D-GS) methods neglect spatiotemporal redundancy in dynamic scenes, resulting in prohibitive memory overhead. This paper introduces the first efficient compression framework for 4D Gaussian lattices, innovatively adapting video compression principles to neural rendering: we design a 3D-anchor-based spatiotemporal prediction module to model inter-frame geometric and appearance correlations; integrate adaptive quantization with context-aware entropy coding to deeply exploit redundancy. Evaluated on synthetic and real-world dynamic scenes, our method achieves 40× and 90× storage compression, respectively—reducing average model size to ~1 MB—while preserving state-of-the-art reconstruction quality and real-time rendering speed. To our knowledge, this is the first end-to-end trainable 4D-GS compression approach achieving high fidelity, high compression ratio, and practical efficiency.

Technology Category

• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Learning on the Edge & Model Compression

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

3D Gaussian Splatting (3DGS) has garnered significant attention due to its superior scene representation fidelity and real-time rendering performance, especially for dynamic 3D scene reconstruction ( extit{i.e.}, 4D reconstruction). However, despite achieving promising results, most existing algorithms overlook the substantial temporal and spatial redundancies inherent in dynamic scenes, leading to prohibitive memory consumption. To address this, we propose P-4DGS, a novel dynamic 3DGS representation for compact 4D scene modeling. Inspired by intra- and inter-frame prediction techniques commonly used in video compression, we first design a 3D anchor point-based spatial-temporal prediction module to fully exploit the spatial-temporal correlations across different 3D Gaussian primitives. Subsequently, we employ an adaptive quantization strategy combined with context-based entropy coding to further reduce the size of the 3D anchor points, thereby achieving enhanced compression efficiency. To evaluate the rate-distortion performance of our proposed P-4DGS in comparison with other dynamic 3DGS representations, we conduct extensive experiments on both synthetic and real-world datasets. Experimental results demonstrate that our approach achieves state-of-the-art reconstruction quality and the fastest rendering speed, with a remarkably low storage footprint (around extbf{1MB} on average), achieving up to extbf{40$ imes$} and extbf{90$ imes$} compression on synthetic and real-world scenes, respectively.

Problem

• Addresses high memory consumption in dynamic 3D scene reconstruction
• Exploits spatial-temporal redundancies for compact 4D scene modeling
• Achieves high compression rates while maintaining reconstruction quality

Innovation

• Predictive 4D Gaussian Splatting with 90x compression
• Spatial-temporal prediction module for Gaussian primitives
• Adaptive quantization with entropy coding for compression