HybridSplat: Fast Reflection-baked Gaussian Tracing using Hybrid Splatting

2025-12-09

AI Summary

To address the slow rendering speed and high memory overhead of 3D Gaussian Splatting in complex reflective scenes, this paper proposes a reflection-aware hybrid splatting framework. Methodologically, it introduces (1) a novel reflection-baked Gaussian tracing mechanism that explicitly models multi-bounce reflection paths; (2) a synergistic optimization combining tile-based Gaussian rasterization with hierarchical, reflection-sensitive culling to jointly reduce computational and memory costs; and (3) an end-to-end pipeline-accelerated architecture. Evaluated on standard reflective benchmarks—including Ref-NeRF and NeRF-Casting—the method achieves approximately 7× rendering acceleration while maintaining state-of-the-art visual quality using only 40% of the Gaussian primitives. This significantly improves both efficiency and fidelity in novel-view synthesis for complex reflective environments.

Technology Category

• Computer Vision: 3D Computer Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Hardware-aware ML

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Data transparency and provenance
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Rendering complex reflection of real-world scenes using 3D Gaussian splatting has been a quite promising solution for photorealistic novel view synthesis, but still faces bottlenecks especially in rendering speed and memory storage. This paper proposes a new Hybrid Splatting(HybridSplat) mechanism for Gaussian primitives. Our key idea is a new reflection-baked Gaussian tracing, which bakes the view-dependent reflection within each Gaussian primitive while rendering the reflection using tile-based Gaussian splatting. Then we integrate the reflective Gaussian primitives with base Gaussian primitives using a unified hybrid splatting framework for high-fidelity scene reconstruction. Moreover, we further introduce a pipeline-level acceleration for the hybrid splatting, and reflection-sensitive Gaussian pruning to reduce the model size, thus achieving much faster rendering speed and lower memory storage while preserving the reflection rendering quality. By extensive evaluation, our HybridSplat accelerates about 7x rendering speed across complex reflective scenes from Ref-NeRF, NeRF-Casting with 4x fewer Gaussian primitives than similar ray-tracing based Gaussian splatting baselines, serving as a new state-of-the-art method especially for complex reflective scenes.

Problem

• Accelerates rendering speed for complex reflective scenes
• Reduces memory storage by pruning Gaussian primitives
• Maintains high-fidelity reflection quality in novel view synthesis

Innovation

• Reflection-baked Gaussian tracing for view-dependent effects
• Unified hybrid splatting framework for high-fidelity reconstruction
• Pipeline acceleration and reflection-sensitive pruning for efficiency