Underwater optical imaging is severely hindered by scattering, but polarization imaging offers the unique dual advantages of descattering and shape-from-polarization (SfP) 3D reconstruction. To exploit these advantages, this paper proposes UD-SfPNet, an underwater descattering shape-from-polarization network that leverages polarization cues for improved 3D surface normal prediction. The framework jointly models polarization-based image descattering and SfP normal estimation in a unified pipeline, avoiding error accumulation from sequential processing and enabling global optimization across both tasks. UD-SfPNet further incorporates a novel color embedding module to enhance geometric consistency by exploiting the relationship between color encodings and surface orientation. A detail enhancement convolution module is also included to better preserve high-frequency geometric details that are lost under scattering. Experiments on the MuS-Polar3D dataset show that the proposed method significantly improves reconstruction accuracy, achieving a mean surface normal angular error of 15.12∘ (the lowest among compared methods). These results confirm the efficacy of combining descattering with polarization-based shape inference, and highlight the practical significance and potential applications of UD-SfPNet for optical 3D imaging in challenging underwater environments. The code is available at https://github.com/WangPuyun/UD-SfPNet.
Figures & tables
Figure 2: Underwater polarization imaging model.
Figure 3: Schematic of the minimal element in the surface microfacet model.
Figure 4: Physical model of polarization-based 3D reconstruction: (a) global view; (b) camera view.
Figure 5: Relation between the degree of polarization ρ and the zenith angle θ for a given refractive index η : (a) specular reflection; (b) diffuse reflection.
Figure 6: Relation between the intensity I and the polarizer rotation angle φ .
Figure 7: Overview of the structured learning framework UD-SfPNet for underwater polarization-based 3D reconstruction.
Figure 8: Color encoding mechanism of baked normal maps.
Figure 9: Color embedding module adapted from DCC [ 31 ] : the pyramid color embedding (PCE) composed of multi-level color embedding modules (left), and the detailed implementation of the color embedding (CE) module (right).
Figure 10: Comparison of error heatmaps across different samples.
Figure 11: Schematic illustration of the DEConv module.
Figure 12: Descattering evaluation using the false match rate of ORB-based registration. Green circles indicate detected keypoints, and red lines indicate incorrect matches.
Metric
PSNR ↑
SSIM ↑
LPIPS ↓
Raw data
30.80
0.9569
0.3830
Descattering result
36.87
0.9745
0.0356
Table 1: Quantitative comparison before and after descattering. Masks are applied to both the ground truth and the evaluated samples to focus on the target region.
Metric
DeepSfP
SfP-wild
TransSfP
Attention U2-Net
DSINE
Proposed
MAE ↓
19.64 ∘
21.64 ∘
20.54 ∘
15.72 ∘
16.94 ∘
15.12 ∘
Table 2: Performance comparison of different baselines on the MuS-Polar3D test set. The metric is mean angular error (MAE, ∘ ); lower is better.
Figure 13: Comparison of different baseline methods on a representative sample (Dragon2). The red numbers in the upper-right corner of each subfigure denote the MAE (°), where lower values indicate better performance.
Figure 14: Comparison of different normal estimation methods on multiple samples under the same turbidity condition (10 ml emulsion). For each group, the results are presented in the order of predicted normal maps, enlarged local details, and angular error heatmaps. The red numbers indicate the MAE (°) of the corresponding method on each sample.
Figure 15: 3D shape integration results from predicted normal maps using a normal integration method. (a) Integrated shapes of different baselines under 10 ml emulsions. (b) Integrated shapes of the proposed method under different scattering conditions.
Model
w/o PPN
w/o DN
w/o PPN & DN
w/o CE
w/o DEConv
Proposed
Mean
16.72 ∘
15.37 ∘
15.56 ∘
15.46 ∘
23.03 ∘
15.12 ∘
Median
15.94 ∘
15.38 ∘
16.09 ∘
15.73 ∘
22.48 ∘
15.21 ∘
Table 3: Ablation results of key sub-networks/modules. The metrics are mean/median normal angular error ( ∘ ); lower is better.
Underwater 3D reconstruction recovers scene geometry and appearance from camera, sonar, and robot-sensor data despite refraction, scattering, and motion.
Reliable feedforward underwater 3D reconstruction remains challenging due to severe light attenuation and backscattering, which degrade visual quality and disrupt feature consistency across views, leading to inaccurate multi-view geometry. To address this issue, we propose WAT3R, a feed-forward framework for reconstructing 3D scenes directly from underwater images. By leveraging degradation adaptation as a geometry-constrained process, WAT3R integrates a lightweight neural adaptation module to flexibly account for these underwater imaging effects, thereby improving multi-view reconstruction quality. Implemented in a single forward pass, WAT3R directly and efficiently outputs pixel-aligned 3D point maps and camera poses from underwater videos, allowing a high-quality underwater 3D reconstruction. Experiments conducted on the FLSea, SQUID, and USOD10K datasets show that our method consistently outperforms state-of-the-art approaches on 3D reconstruction tasks, including multi-view/monocular depth estimation and camera pose estimation.
Jiayi Xu, Jiahao Lu, Ziqiang Zheng +4
The Hong Kong University of Science and Technology · The Chinese University of Hong Kong · Peking University
3D imaging of specular surfaces remains challenging in real-world scenarios, such as in-line inspection or hand-held scanning, requiring fast and accurate measurement of complex geometries. Optical metrology techniques such as deflectometry achieve high accuracy but typically rely on multi-shot acquisition, making them unsuitable for dynamic environments. Fourier-based single-shot approaches alleviate this constraint, yet their performance deteriorates when measuring surfaces with high spatial frequency structure or large curvature. Alternatively, polarimetric 3D imaging in computer vision operates in a single-shot fashion and exhibits robustness to geometric complexity. However, its accuracy is fundamentally limited by the orthographic imaging assumption. In this paper, we propose a physics-informed deep learning framework for single-shot 3D imaging of complex specular surfaces. Polarization cues provide orientation priors that assist in interpreting geometric information encoded by structured illumination. These complementary cues are processed through a dual-encoder architecture with mutual feature modulation, allowing the network to resolve their nonlinear coupling and directly infer surface normals. The proposed method achieves accurate and robust normal estimation in single-shot with fast inference, enabling practical 3D imaging of complex specular surfaces.
Jiazhang Wang, Hyelim Yang, Tianyi Wang +1
Wyant College of Optical Sciences, University of Arizona, Tucson, AZ 85721
We propose Swimm3R, a unified framework that combines medium-aware structure-from-motion (SfM) with Underwater Beta Splatting to address scattering- and attenuation-induced failures in underwater 3D reconstruction. Swimm3R distills in-air geometric priors into a feed-forward backbone and uses a physics head to regress underwater image-formation parameters, camera poses, and restored point clouds. Additionally, we introduce Underwater Beta Splatting, which extends Gaussian splatting with Beta primitives and scattering-aware geometric gradients for stable underwater geometry representation. We further establish the Barbados underwater video dataset to demonstrate the effectiveness of our method in challenging underwater environments. On this dataset, Swimm3R robustly recovers underwater scene structure under challenging scattering conditions, yielding coherent seafloor geometry. Using these predicted point clouds, the proposed Underwater Beta Splatting improves average PSNR by 1.47 dB over WaterSplatting while increasing downstream localization performance by 2.0 and 2.4 percentage points in RRA@15 and RTA@15, respectively.
Minseong Kweon, Junaed Sattar
Department of Computer Science & Engineering and the Minnesota Robotics Institute (MnRI), University of Minnesota–Twin Cities, Minneapolis, MN 55455 USA