3D Gaussian Splatting (3DGS) has emerged as a powerful explicit representation enabling real-time, high-fidelity 3D reconstruction and novel view synthesis. However, its practical use is hindered by the massive memory and computational demands required to store and render millions of Gaussians. These challenges become even more severe in 4D dynamic scenes.

To address these issues, the field of Efficient Gaussian Splatting has rapidly evolved, proposing methods that reduce redundancy while preserving reconstruction quality. This survey provides the first unified overview of efficient 3D and 4D Gaussian Splatting techniques.

For both 3D and 4D settings, we systematically categorize existing methods into two major directions, Parameter Compression and Restructuring Compression, and comprehensively summarize the core ideas and methodological trends within each category. We further cover widely used datasets, evaluation metrics, and representative benchmark comparisons. Finally, we discuss current limitations and outline promising research directions toward scalable, compact, and real-time Gaussian Splatting for both static and dynamic 3D scene representation.

Directly compresses Gaussian attributes without modifying the original 3DGS architecture, eliminating redundancy while preserving rendering quality through five methods: Pruning, Attribute Pruning, Quantization, Entropy Coding, and Structured Compression.

Fundamentally modifies the original 3DGS architecture to achieve efficient scene representation through three strategies: Anchor-based Hierarchical Structure methods using sparse anchors, Neural Network Integration methods replacing representations with neural networks, and Geometric Structure-aware methods exploiting scene geometric properties.

Reduces redundancy in spatio-temporal representations without modifying the rendering architecture through four methods: Gaussian Pruning based on temporal activity, Attribute Pruning removing temporal parameters, Quantization discretizing Gaussian parameters, and Entropy-based Compression exploiting statistical redundancy.

Achieves compact representations by restructuring the underlying architecture through three approaches: Anchor-based Representation modeling deformations via sparse anchors, Canonical Deformable Representation mapping canonical space to time-varying space using deformation fields, and LoD Representation organizing Gaussians into multi-resolution hierarchies.

The left figure presents a bubble chart comparing static 3DGS methods on Mip-NeRF 360, where the x-axis shows perceptual quality (LPIPS ↓), the y-axis shows reconstruction quality (PSNR ↑), and bubble size represents model storage size (MB ↓), demonstrating that various compact designs achieve improved quality with reasonable storage.

The right figure presents a bubble chart comparing dynamic 3DGS methods on N3DV, where the x-axis shows rendering speed (FPS ↑), the y-axis shows reconstruction quality (PSNR ↑), and bubble size represents model storage size (MB ↓), revealing that compact models achieve substantial storage reduction while maintaining high quality and rendering speed.

Despite advances in efficient 3DGS, we identify critical challenges in hardware optimization for resource-constrained devices, long-sequence processing for dynamic scenes, and semantically-aware compression strategies. We propose future research directions including developing generalizable foundation models without per-scene optimization, enabling user-controllable quality-efficiency trade-offs, and enhancing reliability and robustness for safety-critical applications.