We present SSD-GS, a physically-based relighting framework built upon 3D Gaussian Splatting (3DGS) that achieves high-quality reconstruction and photorealistic relighting under novel lighting conditions. In physically-based relighting, accurately modeling light-material interactions is essential for faithful appearance reproduction. However, existing 3DGS-based relighting methods adopt coarse shading decompositions, either modeling only diffuse and specular reflections or relying on neural networks to approximate shadows and scattering. This leads to limited fidelity and poor physical interpretability, particularly for anisotropic metals and translucent materials. To address these limitations, SSD-GS decomposes reflectance into four components: diffuse, specular, shadow, and subsurface scattering. We introduce a learnable dipole-based scattering module for subsurface transport, an occlusion-aware shadow formulation that integrates visibility estimates with a refinement network, and an enhanced specular component with an anisotropic Fresnel-based model. Through progressive integration of all components during training, SSD-GS effectively disentangles lighting and material properties, even for unseen illumination conditions, as demonstrated on the challenging OLAT dataset. Experiments demonstrate superior quantitative and perceptual relighting quality compared to prior methods and pave the way for downstream tasks, including controllable light source editing and interactive scene relighting. The source code is available at: https://github.com/irisfreesiri/SSD-GS.
Figures & tables
Figure 1: Overview of the proposed SSD-GS pipeline. Our method incorporates four physically inspired reflectance terms: diffuse, specular, shadow, and subsurface scattering, to model realistic light–material interactions. These components are progressively introduced during training, allowing the network to gradually disentangle complex illumination effects and improve relighting fidelity under unseen lighting conditions.
Figure 2: Relighting results from our SSD-GS pipeline. The same Bunny view under two different lighting conditions from the SSS-GS synthetic dataset ( Dihlmann et al., 2025 ) .
Figure 3: Shadow pipeline visualization. For each scene, we show per-ray transmittance vi , coarse visibility v^g , the refined shadow S(x) , and the final rendered result with metrics. The progression illustrates how continuous transmittance yields smooth, geometry-aware soft shadows.
Table 1: Quantitative comparison results. The best/second-best results are colored in red / orange .
Figure 4: Qualitative comparison on real datasets from NRHints ( Zeng et al., 2023 ) . It presents relighting results on the test set under novel lighting. The best/second-best results (based on PSNR) are highlighted in red / orange .
Figure 5: Visualization of reconstructed components under different reflectance decompositions and training schedules on the Pixiu scene from the NRHints real dataset. Top: six reflectance compositions (Comp. A-F); Bottom: four training schedules (Sched. H-K).
Table 2: Ablation study results on the real scene Pixiu .
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 6: Illustration of the progressive training schedule on the pixiu scene from the NRHints real dataset ( Zeng et al., 2023 ) . The model is trained for a total of 100K iterations.
Figure 7: Illustration of the progressive training schedule on the bunny scene from the SSS-GS synthetic dataset ( Dihlmann et al., 2025 ) . The model is trained for a total of 60K iterations.
Figure 8: Qualitative comparison of relighting results on novel test-time lighting from synthetic datasets provided by GS3 ( Bi et al., 2024 ) . Note that the rendered view of RNG’s results on the Drums and Lego scenes are obtained using their official code trained using a set of white-background images.
Table 3: Quantitative comparison with SSS-GS ( Dihlmann et al., 2025 ) and KiloOSF ( Yu et al., 2022 ) on the synthetic datasets provided by SSS-GS, trained for 60K iterations. The best and second-best results (based on PSNR) are highlighted in red and orange , respectively.
Figure 9: Qualitative comparison on synthetic datasets from SSS-GS ( Dihlmann et al., 2025 ) . The left column shows reconstruction results on training views, while the right column presents relighting results on novel test-time lighting. The best and second-best results (based on PSNR) are highlighted in red and orange , respectively.
Figure 10: Qualitative relighting results on both real and synthetic datasets. Each scene is rendered from the same camera viewpoint under six novel lighting conditions.
Figure 11: Qualitative relighting comparison on the Translucent scene. Each row shows renderings from the same camera viewpoint under four novel lighting conditions. Our method is compared against RNG ( Fan et al., 2025 ) and GS 3 ( Bi et al., 2024 ) , demonstrating superior fidelity in reproducing light–material interactions.
Dataset
NRHints
GS 3
SSS-GS
Average
Scenes
Pixiu
Fish
Translucent
FurBall
Lego
Hotdog
Bunny
Dragon
PSNR ↑
A: Diff
Train
20.2869
23.7003
16.6152
18.5678
20.6157
18.4717
21.3548
27.7851
20.9247
Test
20.1878
24.6146
15.3146
17.4010
17.6430
16.4279
21.3013
27.7071
20.0747
B: D + S
Train
20.7321
24.8949
16.6980
18.3871
20.6891
18.5180
21.4677
27.9589
21.1682
Test
20.5692
25.3627
15.3200
17.1781
17.6290
16.4429
21.2236
27.6713
20.1746
Appendix
Table 4: Ablation study results on both real and synthetic datasets. The best/second-best results are colored in red / orange .
Figure 12: Visualization of reconstructed components under different reflectance decompositions and training schedules on the Translucent scene from the GS 3 synthetic dataset. Top: six reflectance compositions (Comp. A-F); Bottom: four training schedules (Sched. H-K).
Dataset
NRHints
GS 3
SSS-GS
Average
Scenes
Pixiu
Fish
FurScene
Translucent
FurBall
Lego
Hotdog
Bunny
Dragon
PSNR ↑
Shadow-on-SSS
Train
32.1850
31.7531
31.6750
30.8066
35.4219
31.1002
32.1094
35.9068
37.3346
33.1436
Test
30.9756
31.1206
30.6575
30.0740
34.9420
30.3487
31.5272
33.6508
35.8476
32.1271
Ours
Train
33.6065
32.0748
31.7846
32.6058
35.4793
31.1434
32.4901
40.7672
39.3646
34.3685
Test
31.1213
31.1646
30.7349
32.3919
35.1639
30.4664
32.1330
37.2270
36.6325
33.0039
Appendix
Table 5: Quantitative comparison when applying the shadow term to the SSS component.
Figure 13: Qualitative comparison for the Shadow–SSS interaction ablation study. Applying the shadow term to the SSS component leads to over-darkened scattering, loss of back-lit translucency, and reduced soft shading. Our formulation preserves translucent appearance while maintaining consistent shadow behavior.
Table 6: Quantitative comparison results with expanded relighting baselines: TensoIR and R3DG. The best results are colored in red .
Figure 14: Comparison with screen-space shadow baselines. Top row: reference renderings. Second row: shadows produced by our method. Third and fourth rows: shadows from GS 3 ( Bi et al., 2024 ) and RNG ( Fan et al., 2025 ) , which both rely on screen-space opacity accumulation.
We present VolS-GS, a relightable Gaussian splatting framework that reconstructs objects from one-light-at-a-time (OLAT) captures and renders them under novel lighting and viewpoints. Relightable Gaussian Splatting methods typically model appearance independently at each primitive, which makes non-local effects difficult to represent. This limitation is particularly apparent for subsurface scattering, where light entering the object at one location can emerge at another. Rather than modeling this effect solely with a neural network or a local kernel at each primitive, we use the spatial support of the Gaussian scene as the domain of a differentiable finite-volume transport solver, so that light can propagate through the object's interior. A small network predicts scattering and absorption coefficients for each Gaussian, and the solve redistributes incident light through the resulting field. The coefficients are fit to images rather than measured, so the solve supplies a transport-shaped path for aggregating per-primitive appearance, not a measurement of the material. To keep the learned shadow and specular terms from taking over the other components, our shadow term is predicted from visibility together with the transmittances and the scattering the solve produces, and a regularizer suppresses specular highlights in regions the shadow term predicts to be unlit. Experiments on three OLAT benchmarks show that VolS-GS consistently improves relighting quality on held-out lights and views.
We address inverse rendering with 3D Gaussian Splatting (3DGS) by introducing explicit visibility modeling for relighting as well as material and light decomposition. Our approach augments the standard rasterized 3DGS pipeline with a sparse visibility oracle that estimates transmittance-weighted occlusion for Gaussian primitives from sampled lighting directions. This oracle is implemented using a point-based ray tracer accelerated by a bounding volume hierarchy (BVH) and queried sparsely within the inverse-rendering objectives, while the original 3DGS loop is retained for training efficiency, interactive visualization, and tooling compatibility. Shading follows a physically based rendering (PBR) formulation with a principled microfacet bidirectional reflectance distribution function (BRDF) and low-resolution high-dynamic-range (HDR) environment maps. We evaluate our method on the TensoIR synthetic dataset against baselines based on neural radiance fields (NeRFs), tensor representations, and 3DGS. The results show competitive performance and improvements over several earlier baselines for novel-view synthesis (NVS), albedo reconstruction, and relighting. Ablations indicate that explicit visibility modeling is an important contributor to relighting and material-illumination disentanglement, while pure PBR supervision can favor appearance fitting in NVS. Overall, the findings suggest that sparse, point-based visibility queries effectively complement rasterized splatting for inverse rendering while preserving the practical advantages of the 3DGS ecosystem.
Sven Kluge, Oliver Staadt
University of Rostock, Albert-Einstein-Straße 22, 18059 Rostock, Germany
3D Gaussian Splatting (3DGS) has established itself as a leading technique for 3D reconstruction and novel view synthesis of static scenes, achieving outstanding rendering quality and fast training. However, the method does not explicitly model the scene illumination, making it unsuitable for relighting tasks. Furthermore, 3DGS struggles to reconstruct scenes captured in the wild by unconstrained photo collections featuring changing lighting conditions. In this paper, we present R3GW, a novel method that learns a relightable 3DGS representation of an outdoor scene captured in the wild. Our approach separates the scene into a relightable foreground and a non-reflective background (the sky), using two distinct sets of Gaussians. R3GW models view-dependent lighting effects in the foreground reflections by combining Physically Based Rendering with the 3DGS scene representation in a varying illumination setting. We evaluate our method quantitatively and qualitatively on the NeRF-OSR dataset, offering state-of-the-art performance and enhanced support for physically-based relighting of unconstrained scenes. Our method synthesizes photorealistic novel views under arbitrary illumination conditions. Additionally, our representation of the sky mitigates depth reconstruction artifacts, improving rendering quality at the sky-foreground boundary
Margherita Lea Corona, Wieland Morgenstern, Peter Eisert +1
Fraunhofer Heinrich Hertz Institute, Berlin, Germany · Humboldt Universität zu Berlin, Berlin, Germany