We present ReflFlow, a novel framework for high-fidelity rendering of dynamic specular scenes by addressing two key challenges: precise reflection direction estimation and physically accurate modeling. To achieve this, we propose a Residual Material-Augmented 2D Gaussian Splatting representation that models dynamic geometry and material properties, allowing accurate reflection ray computation. Furthermore, we introduce a Dynamic Environment Gaussian and a hybrid rendering pipeline that decomposes rendering into diffuse and specular components, enabling physically informed specular synthesis via rasterization and ray tracing. Finally, we devise a coarse-to-fine training strategy to improve optimization stability and promote physically meaningful decomposition. Extensive experiments on dynamic scene benchmarks demonstrate that ReflFlow outperforms prior methods quantitatively and qualitatively, producing sharper and more realistic specular reflections in complex dynamic environments.
Figures & tables
Figure 1: ReflFlow shows the sharpest and most photorealistic specular details among all compared approaches. PSNR ↑ and SSIM ↑ should be as high as possible. The performance shown in the figure corresponds to the Plate scene. Please zoom in for a clearer view.
Figure 2: Overview of ReflFlow. (a) For a dynamic specular scene, at each timestamp, a viewing ray is traced from the camera. After intersecting with the main content, it reflects off the surface based on the surface normal. The resulting reflection ray then intersects with the dynamic environment. (b) To render such a scene, we use rasterization to compute the diffuse color of the main content and employ a ray tracer to compute the specular color via the reflection ray. Finally, the diffuse and specular components are blended to the final color.
Figure 3: Qualitative Comparison Results on the NeRF-DS Dataset. Our method significantly improves the visual quality of dynamic specular reconstruction compared to previous approaches. In particular, it produces sharper details and fewer artifacts in specular regions, demonstrating enhanced fidelity in modeling dynamic reflections. Please zoom in for more details.
As
Basin
Bell
Cup
Method
PSNR↑
SSIM↑
LPIPS↓
PSNR↑
SSIM↑
LPIPS↓
PSNR↑
SSIM↑
LPIPS↓
PSNR↑
SSIM↑
LPIPS↓
Deformable 3DGS ( Yang et al., 2024b )
26.04
0.8805
0.1850
19.53
0.7855
0.1924
23.96
0.7945
0.2767
24.49
0.8822
0.1658
4DGS ( Yang et al., 2024c )
24.85
0.8632
0.2038
19.26
0.7670
0.2196
22.86
0.8015
0.2061
23.82
0.8695
0.1792
GaussianShader ( Jiang et al., 2024 )
21.89
0.7739
0.3620
17.79
0.6670
0.4187
20.69
0.8169
0.3024
20.40
0.7437
0.3385
GS-IR ( Liang et al., 2024 )
21.58
0.8033
0.3033
18.06
0.7248
0.3135
20.66
0.7829
0.2603
20.34
0.8193
0.2719
NeRF-DS ( Yan et al., 2023 )
25.34
0.8803
0.2150
20.23
0.8053
0.2508
22.57
0.7811
0.2921
24.51
0.8802
0.1707
Table 1: Quantitative comparison on the NeRF-DS ( Yan et al., 2023 ) dataset. We report the average PSNR, SSIM, and LPIPS (VGG) across seven scenes. The best , the second best , and the third best results are denoted by red, orange, yellow.
Method
PSNR↑
SSIM↑
LPIPS↓
General dynamic reconstruction methods
Deformable 3DGS ( Yang et al., 2024b )
22.78
0.6201
0.3380
4DGS ( Yang et al., 2024c )
24.89
0.6781
0.3422
HyperNeRF ( Park et al., 2021b )
23.11
0.6387
0.3968
Specular reconstruction methods
NeRF-DS ( Yan et al., 2023 )
23.65
0.6405
0.3972
Table 2: Quantitative comparison on HyperNeRF ( Park et al., 2021b ) . Best and second best results are highlighted.
Figure 4: Qualitative comparison of ablation study on different components. ” + ” denotes the incremental addition of each component to the previous configuration, starting from the base model.
FθG
Fθenv
Lnorm
Ltc-norm
PSNR↑
SSIM↑
LPIPS↓
16.29
0.6655
0.4126
✓
18.97
0.7793
0.2389
✓
✓
20.25
0.8201
0.2313
✓
✓
✓
22.12
0.8441
0.2149
✓
✓
✓
✓
25.23
0.8855
0.1536
Table 3: Ablation studies on different components.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 5: More results on NeRF-DS datasets. Our method can recover fine-grained specular details in dynamic specular reconstruction.
Figure 6: Visualized our rendering images, normal maps, and depth maps.
Method
GPU
Iterations
Training Time
FPS↑
SpectroMotion ( Fan et al., 2024 )
RTX 6000 Ada
40,000
1.1 hours
33
Ours
RTX 6000 Ada
60,000
2.8 hours
32
Appendix
Table 4: Efficiency comparison with SpectroMotion on NVIDIA RTX 6000 Ada. Our method achieves comparable inference FPS while providing superior reconstruction quality.
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
While novel view synthesis (NVS) for dynamic scenes has seen significant progress, reconstructing temporally consistent geometric surfaces remains a challenge. Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) offer powerful dynamic scene rendering capabilities; however, relying solely on photometric optimization often leads to geometric ambiguities. This results in discontinuous surfaces, severe artifacts, and broken surfaces over time. To address these limitations, we present DySurface, a novel framework that bridges the effectiveness of explicit Gaussians with the geometric fidelity of implicit Signed Distance Functions (SDFs) in dynamic scenes. Our approach tackles the structural discrepancy between the forward deformation of 3DGS (canonical→dynamic) and the backward deformation required for volumetric SDF rendering (dynamic→canonical). Specifically, we propose the VoxGS-DSDF branch that leverages deformed Gaussians to construct a dynamic sparse voxel grid, providing explicit geometric guidance to the implicit SDF field. This explicit anchoring effectively regularizes the volumetric rendering process, significantly improving surface reconstruction quality, with watertight boundaries and detailed representations. Quantitative and qualitative experiments demonstrate that DySurface significantly outperforms state-of-the-art baselines in geometric accuracy while maintaining competitive rendering performance.
While novel view synthesis methods like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have shown impressive performance on common scenes, they often struggle with capturing view-dependent specular effects, particularly near-field reflections. Existing methods utilize environment maps, which only represent lights from infinity, or environment Gaussians, which rely solely on indirect supervision from reflections, are insufficient for accurately modeling such effects. Here, we propose a framework based on 2D Gaussian Splatting (2DGS) that leverages directly observable environment information in input images. Our approach also integrates a hybrid reflection appearance representation and view-dependent specular modeling. Experiments on public benchmarks and a newly curated near-field reflection dataset demonstrate that our method achieves superior rendering quality and geometric accuracy.
Yiming Liang, Tianyu Xiao, Hiroshi Ishikawa
Department of Computer Science and Communications Engineering, Waseda University, Japan