PolarGuide-GSDR: 3D Gaussian Splatting Driven by Polarization Priors and Deferred Reflection for Real-World Reflective Scenes

TL;DR

PolarGuide-GSDR addresses the challenge of real-world reflective scene reconstruction by embedding polarization priors into 3D Gaussian Splatting with a deferred reflection mechanism. It introduces a bidirectional coupling: polarization-based specular/diffuse separation provides priors for 3DGS normals, while refined polarization cues guide the 3DGS SH-based reflectance representation. The framework resolves polarization normal ambiguities with 3DGS priors and uses a deferred reflection supervisory loop to tighten geometric accuracy, all without environment maps. Experiments on public and self-collected datasets demonstrate state-of-the-art specular reconstruction, normal estimation, and novel-view synthesis, while maintaining real-time rendering performance.

Abstract

Polarization-aware Neural Radiance Fields (NeRF) enable novel view synthesis of specular-reflection scenes but face challenges in slow training, inefficient rendering, and strong dependencies on material/viewpoint assumptions. However, 3D Gaussian Splatting (3DGS) enables real-time rendering yet struggles with accurate reflection reconstruction from reflection-geometry entanglement, adding a deferred reflection module introduces environment map dependence. We address these limitations by proposing PolarGuide-GSDR, a polarization-forward-guided paradigm establishing a bidirectional coupling mechanism between polarization and 3DGS: first 3DGS's geometric priors are leveraged to resolve polarization ambiguity, and then the refined polarization information cues are used to guide 3DGS's normal and spherical harmonic representation. This process achieves high-fidelity reflection separation and full-scene reconstruction without requiring environment maps or restrictive material assumptions. We demonstrate on public and self-collected datasets that PolarGuide-GSDR achieves state-of-the-art performance in specular reconstruction, normal estimation, and novel view synthesis, all while maintaining real-time rendering capabilities. To our knowledge, this is the first framework embedding polarization priors directly into 3DGS optimization, yielding superior interpretability and real-time performance for modeling complex reflective scenes.

Figures (8)

• Figure 1: Novel view synthesis results with specular reflections. From left to right: ground-truth, GNeRF yang2024gnerp, 3DGS kerbl2023gaussian, 3DGS-DR ye2024deferredreflection, Ref-GS zhang2025refgs, and ours. By introducing polarization information for guidance and supervision, our method achieves higher-quality reflection reconstruction.
• Figure 2: Overview of Polarguide-3DGS. First, polarization images are input to the system. The polarization information calculation module computes DoLP and AoLP, and extracts four-channel polarization images. Then, the Polarization Model-Based Specular and Diffuse Reflection Separation generates polarized specular and diffuse images as priors. And the initial polarization surface normal prior is estimated from DoLP and AoLP, then refined by the Polarization Normal Ambiguity-Correction Strategy Guided by 3DGS Priors to resolve normal estimation ambiguities. Finally, the polarized Specular and Diffuse Reflection images, refined normals, and RGB ground truth jointly guide and supervise the pre-trained 3DGS model to enhance reflectance and geometry learning. During rendering, the estimated reflection strength map blends specular and diffuse outputs to produce realistic reflections.
• Figure 3: Separation results of specular and diffuse components based on polarized images.
• Figure 4: Comparison of Test Views Across Different Indoor and Outdoor Scenes (Note: Surface normals in GT are derived from polarization data, not measured by scanning, and are included to demonstrate the effectiveness of polarization-based normal extraction.)
• Figure 5: Comparison of Test Views on Public Datasets