PolGS++: Physically-Guided Polarimetric Gaussian Splatting for Fast Reflective Surface Reconstruction
Authors: Yufei Han, Chu Zhou, Youwei Lyu, Qi Chen, Si Li, Boxin Shi, Yunpeng Jia, Heng Guo, Zhanyu Ma
Organizations: Beijing University of Posts and Telecommunications, China · National Institute of Informatics, Japan · vivo BlueImage Lab, vivo Mobile Communication Co., Ltd., Shanghai, China · State Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University, China · National Engineering Research Center of Visual Technology, School of Computer Science, Peking University, China
Accurate reconstruction of reflective surfaces remains a fundamental challenge in computer vision, with broad applications in real-time virtual reality and digital content creation. Although 3D Gaussian Splatting (3DGS) enables efficient novel-view rendering with explicit representations, its performance on reflective surfaces still lags behind implicit neural methods, especially in recovering fine geometry and surface normals. To address this gap, we propose PolGS++, a physically-guided polarimetric Gaussian Splatting framework for fast reflective surface reconstruction. Specifically, we integrate a polarized BRDF (pBRDF) model into 3DGS to explicitly decouple diffuse and specular components, providing physically grounded reflectance modeling and stronger geometric cues for reflective surface recovery. Furthermore, we introduce a depth-guided visibility mask acquisition mechanism that enables angle-of-polarization (AoP)-based tangent-space consistency constraints in Gaussian Splatting without costly ray-tracing intersections. This physically guided design improves reconstruction quality and efficiency, requiring only about 10 minutes of training. Extensive experiments on both synthetic and real-world datasets validate the effectiveness of our method.
Figures & tables
Method
Input
Reflective
Type
Accuracy
Time
NeRO ( Liu et al., 2023 )
RGB Images
✓
SDF
high
8h
MVAS ( Cao et al., 2023 )
Azimuths
✓
SDF
medium
11h
PANDORA ( Dave et al., 2022 )
Pol. Images
✓
SDF
medium
10h
NeRSP ( Han et al., 2024 )
Pol. Images
✓
SDF
high
10h
PISR ( Chen et al., 2024b )
Pol. Images
✓
SDF
medium
0.5h
GS-ROR ( Zhu et al., 2025 )
RGB Images
✓
SDF+GS
medium
1.5h
Table 1: Comparison of different methods in reflective surface reconstruction on real-world datasets ( Dave et al., 2022 ; Chen et al., 2024b ; Han et al., 2024 ) with the image size 512∗612 .
Figure 2: Comparison between SDF-based and 3DGS-based method on geometry representation. ( Top ) The surface normal of a point has a strong relationship with its opacity in NeuS ( Wang et al., 2021 ) . ( Bottom ) The surface normal of a point has no direct dependence on its opacity in Gaussian Surfels ( Dai et al., 2024 ) .
Figure 3: Pipeline of PolGS++. We re-rendered Stokes vectors s^ by using the diffuse color C from 3DGS and specular color Lr from Cubemap encoder module, which is supervised by the ground truth Stokes information. Tangent-space consistency constrain is applied to the surface normal estimated by 3DGS with the depth-guided visibility mask and AoP information.
Figure 4: Comparison of visibility mask acquisition strategies between SDF-based methods and our method based on Gaussian Splatting.
Figure 5: Visualization of visibility masks of one current view in SDF-based networks and in PolGS++.
Hedgehog
Squirrel
Snail
David
Mean
Method
MAE
CD
MAE
CD
MAE
CD
MAE
CD
MAE
CD
time
PANDORA ( Dave et al., 2022 )
9.41
9.50
10.85
5.88
8.08
10.97
14.75
4.88
10.77
7.81
10 h
NeRSP ( Han et al., 2024 )
8.94
6.57
8.23
3.02
5.56
3.72
15.38
4.18
9.53
4.37
10 h
MVAS ( Cao et al., 2023 )
4.30
4.22
6.10
3.73
3.30
7.87
8.47
3.21
5.54
4.76
11 h
GS-ROR ( Zhu et al., 2025 )
9.72
4.45
9.25
4.20
7.24
5.16
14.02
4.14
10.06
4.49
1.5 h
NeRO ( Liu et al., 2023 )
3.40
3.69
3.55
1.86
2.67
3.71
7.64
2.88
4.32
3.04
8 h
Table 2: Comparisons on synthetic dataset (SMVP3D ( Han et al., 2024 ) ) evaluated by mean angular error (MAE) ( ↓ ) with degree and Chamfer distance (CD) ( ↓ ) in millimeter (mm), respectively. Best and second results in SDF-based and 3DGS-based methods (except PolGS ( Han et al., 2025 ) ) are highlighted as 1st and 2nd . The time consumed is shown on the far right side of the table.
RMVP3D ( Han et al., 2024 )
PISR ( Chen et al., 2024b )
Shisa
Frog
L-Rabbit
S-Rabbit
Mean
Method
MAE
CD
MAE
CD
MAE
CD
MAE
CD
MAE
CD
time
PANDORA ( Dave et al., 2022 )
12.93
11.29
15.86
7.88
13.47
16.53
11.45
15.39
13.43
12.77
10 h
NeRSP ( Han et al., 2024 )
10.79
7.39
15.62
6.68
13.33
7.43
15.75
10.27
13.87
7.94
10 h
MVAS ( Cao et al., 2023 )
8.56
9.28
17.63
7.00
11.86
13.82
12.81
12.08
12.72
10.55
11 h
PISR ( Chen et al., 2024b )
8.23
6.71
16.79
5.89
11.23
10.31
12.18
8.49
12.11
9.20
0.5 h
Table 3: Comparisons on real-world datasets (RMVP3D ( Han et al., 2024 ) and PISR ( Chen et al., 2024b ) ) evaluated by mean angular error (MAE) ( ↓ ) with degree and Chamfer distance (CD) ( ↓ ) in millimeter (mm), respectively. Best and second results in SDF-based and 3DGS-based methods (except PolGS ( Han et al., 2025 ) ) are highlighted as 1st and 2nd . The time consumed is shown on the far right side of the table.
Figure 6: Qualitative comparisons on surface normal in SMVP3D ( Han et al., 2024 ) , where our method can outperform existing methods based on the same representation and achieves comparable results with SDF-based methods.
Figure 7: Qualitative comparison shape on PANDORA ( Dave et al., 2022 ) , RMVP3D ( Han et al., 2024 ) and PISR ( Chen et al., 2024b ) , where PolGS produces similar quality of reconstruction mesh compared to SDF-based methods.
Figure 8: Quantitative ablation results on SMVP3D ( Han et al., 2024 ) , where we evaluate the effect of polarimetric supervision under different input views.
Figure 9: Qualitative ablation results on PISR ( Chen et al., 2024b ) , where we compare the reconstruction quality with and without polarimetric supervision.
Figure 10: Ablation results on the visibility mask threshold.
Figure 11: Separation of diffuse and specular components with PANDORA ( Dave et al., 2022 ) , 3DGS-DR ( Ye et al., 2024 ) and PolGS++.
Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated great success in modeling reflective 3D objects and their interaction with the environment via deferred rendering (DR). However, existing methods often struggle with correctly reconstructing physical attributes such as albedo and reflectance, and therefore they do not support high-fidelity relighting. Observing that this limitation stems from the lack of shape and material information in RGB images, we present PhyGaP, a physically-grounded 3DGS method that leverages polarization cues to facilitate precise reflection decomposition and visually consistent relighting of reconstructed objects. Specifically, we design a polarimetric deferred rendering (PolarDR) process to model polarization by reflection, and a self-occlusion-aware environment map building technique (GridMap) to resolve indirect lighting of non-convex objects. We validate on multiple synthetic and real-world scenes, including those featuring only partial polarization cues, that PhyGaP not only excels in reconstructing the appearance and surface normal of reflective 3D objects (~2 dB in PSNR and 45.7% in Cosine Distance better than existing RGB-based methods on average), but also achieves state-of-the-art inverse rendering and relighting capability. Our code will be released soon.
The reflective appearance, especially strong and typically near-field specular reflections, poses a fundamental challenge for accurate surface reconstruction and novel view synthesis. Existing Gaussian splatting methods either fail to model near-field specular reflections or rely on explicit ray tracing at substantial computational cost. We present Ref-DGS, a reflective dual Gaussian splatting framework that addresses this trade-off by decoupling surface reconstruction from specular reflection within an efficient rasterization-based pipeline. Ref-DGS introduces a dual Gaussian scene representation consisting of geometry Gaussians and complementary local reflection Gaussians that capture near-field specular interactions without explicit ray tracing, along with a global environment reflection field for modeling far-field specular reflections. To predict specular radiance, we further propose a lightweight, physically-aware specular adaptive mixing shader that fuses global and local specular features. Experiments demonstrate that Ref-DGS achieves state-of-the-art performance on reflective scenes while training substantially faster than ray-based Gaussian methods.
Ningjing Fan, Yiqun Wang, Dong-Ming Yan +1
Chongqing University, China · MAIS, Institute of Automation, Chinese Academy of Sciences and UCAS, China · King Abdullah University of Science and Technology (KAUST), Saudi Arabia
In recent years, 3D Gaussian splatting (3DGS) has achieved remarkable progress in novel view synthesis. However, accurately reconstructing glossy surfaces under complex illumination remains challenging, particularly in scenes with strong specular reflections and multi-surface interreflections. To address this issue, we propose SSR-GS, a specular reflection modeling framework for glossy surface reconstruction. Specifically, we introduce a prefiltered Mip-Cubemap to model direct specular reflections efficiently, and propose an IndiASG module to capture indirect specular reflections. Furthermore, we design Visual Geometry Priors (VGP) that couple a reflection-aware visual prior via a reflection score (RS) to downweight the photometric loss contribution of reflection-dominated regions, with geometry priors derived from VGGT, including progressively decayed depth supervision and transformed normal constraints. Extensive experiments on both synthetic and real-world datasets demonstrate that SSR-GS achieves state-of-the-art performance in glossy surface reconstruction.