pith. sign in

arxiv: 2606.15648 · v1 · pith:2GZD3QNOnew · submitted 2026-06-14 · 💻 cs.CV

Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement

Haochen Hu , Yanrui Bin , Zhengyan Zhang , Minchen Wei , Chih-yung Wen , Bing Wang This is my paper

Pith reviewed 2026-06-27 03:58 UTC · model grok-4.3

classification 💻 cs.CV
keywords underwater image enhancementtransfer learningphysics-based decompositionimage restorationcross-domain priorslabel-free learningcomputer vision
0
0 comments X

The pith

Underwater image enhancement works without paired labels by splitting the problem into physics steps and supervising each with priors transferred from other vision tasks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that underwater image enhancement can be performed effectively without any paired noisy or true labels by first decomposing the degradation process according to underwater physics into three distinct operations. It then applies supervision drawn from priors in unrelated vision tasks to each operation separately. This approach is presented as theoretically grounded because the decomposition aligns with physical effects and avoids reliance on imperfect dataset labels. A sympathetic reader would care because true underwater ground truth is hard to obtain, and prior methods suffer from noisy pseudo-labels that limit performance.

Core claim

The central claim is that dividing underwater image enhancement into global color correction, haze removal, and background noise suppression, then solving each step with cross-domain priors transferred from other vision tasks, produces a label-free method that reaches state-of-the-art results on the UIE task and improves downstream vision performance.

What carries the argument

Physics-aligned decomposition of underwater degradation into global color correction, haze removal, and background noise suppression, supervised at each stage by transferred priors from other vision tasks.

If this is right

  • The method achieves state-of-the-art quantitative and qualitative performance on standard UIE benchmarks.
  • Enhanced images improve accuracy on downstream tasks such as object detection or segmentation compared with benchmark UIE methods.
  • The approach requires no paired underwater labels during training.
  • The physics decomposition supplies theoretical soundness to the overall pipeline.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same decomposition-plus-transfer pattern could be tested on other media-specific degradations where physical models exist but paired data do not.
  • Success would imply that many restoration problems currently limited by label noise could be reframed as sequences of simpler, cross-supervised sub-tasks.
  • If the priors transfer reliably, the method may reduce the need for large domain-specific datasets in underwater and similar imaging settings.

Load-bearing premise

The three physical steps capture the dominant underwater degradations and priors from other domains supply useful supervision without needing any domain-specific adaptation or fine-tuning.

What would settle it

Quantitative comparison on a dataset of real underwater images that also possess corresponding clean reference images captured under controlled conditions, measuring whether the method's output metrics exceed those of label-dependent baselines.

Figures

Figures reproduced from arXiv: 2606.15648 by Bing Wang, Chih-Yung Wen, Haochen Hu, Minchen Wei, Yanrui Bin, Zhengyan Zhang.

Figure 1
Figure 1. Figure 1: Example of the noisy labels in the previous paired dataset UIEB[22]. The first, second, and third row are the raw underwater images, labels in UIEB that still suffer low contrast and color bias, and the enhanced results of ours, respectively. The whole process is as follows: according to the Under￾water Image Formation Model (UIFM), we first design a physics-based decomposition strategy, where the UIE is d… view at source ↗
Figure 2
Figure 2. Figure 2: The Concept of our proposal. Instead of supervised learning by noisy labels, we resort to knowledge from relevant research fields for transfer learning via physics-based task decomposition. To the best of our knowledge, this is the first time that multiple priors from relevant vision tasks are utilized for the cross-supervision of UIE. This transfer learning-based strategy as shown in Fig.2 provides a new … view at source ↗
Figure 3
Figure 3. Figure 3: The limitation of finetuning diffusion model by fully synthetic underwater images. (a), (b), (c), (d) are the raw underwater images, enhanced results by [13], the intermediate enhanced result by the global color correction, and the final enhanced result of our overall proposal, respectively. target images by progressively denoising the latent space representation with added noise. It has been widely adopte… view at source ↗
Figure 4
Figure 4. Figure 4: Framework of the Proposal. in some results. This is because the UIFM is a simplified version of the complex real underwater physics. As shown in Fig.3(b), PFUSIE [13] struggles to obtain clear images that preserve the richness and correctness of the color pattern. Therefore, instead of training a UIE model supervised by real world images with noisy labels or by synthetic images based on simplified UIFM, we… view at source ↗
Figure 5
Figure 5. Figure 5: Step 1: Global Color Correction. to generate images with a continuous spectrum of visual clarity, ranging from clear to heavily blurred. In addition, the parameter range is expanded from several discrete values to a continuous range [13] to ensure that the proposal can be adopted under various color biases. Therefore, the general￾izability of the model is guaranteed. In summary, the priors used for transfe… view at source ↗
Figure 6
Figure 6. Figure 6: Step II: Hazy Removal based on Real-world Dense Haze Dataset. Semantic Perception by DINOv3 SAM 2 Semantic Diff. 𝐹 ௗ௜௙௙ in the Feature Space Coarse Haze Semantic Feature 𝐹 ௛ and 𝐹 ௖ AutoMask Point-based Prompt Image 𝐹 ௗ௜௙௙ = 1 − cos 𝐹 ௛ ⋅ 𝐹௖ Automatic Prompt Underwater Image Cos. Projection as Semantic Similarity Low High • Foreground • Background Segmentation on Haze Image [PITH_FULL_IMAGE:figures/full_f… view at source ↗
Figure 7
Figure 7. Figure 7: The process of Automatic Foreground-Background Masking based on Semantic foundation models. The semantic features are colorized after dimensionality reduction by Principal Component Analysis(PCA), while the semantic difference image is normalized to 0-1 and then colorized. training strategy is also proposed: 1) if the foreground segmentation is correct (no need to fully segmented, but the labeled foregroun… view at source ↗
Figure 8
Figure 8. Figure 8: Examples of the final end-to-end training with background noise suppression conducted. Columns (a), (b), (c) are the raw underwater images, enhanced results after global color correction, and enhanced results after haze removal, respectively. Columns (d) and (e) denote the segmentation masks and corresponding results of end-to-end training. Low-dimension Latent Space ℰ 𝐷ఏ ℰ Sampled Noise concat Learnable D… view at source ↗
Figure 9
Figure 9. Figure 9: The pipeline of fine-tuning diffusion model in the latent space. 𝐱0 , 𝐱 are the expected enhanced images and the correspond￾ing degraded images, respectively. In the forward diffusion process, the model first adds a certain noise pattern to the data to destroy its structure, then it learns to reverse the process by denoising the initial noisy input to reconstruct the original data under conditioning input … view at source ↗
Figure 10
Figure 10. Figure 10: Quantitative results on three datasets. Row 1-4, 5-9, 10-15 are sampled from dataset EUVP, LSUI and our self-collected P2UIE_1473, respectively. Column (a) represents the raw underwater images, while columns (b)-(k) are the enhanced results of previous works CCL [26], CE-VAE [35], DPF [31], UIE-CLIP [3], HCLR [55], PUGAN [5], Semi-UIR [14], SSUIE [33], PFUSIE [13], WFDiff [53]. Column (l) illustrates the … view at source ↗
Figure 11
Figure 11. Figure 11: An overview of the raw images ((a),(b)), previous noisy labels ((c),(d)), and the enhanced results of the proposed P2UIE ((e),(f)). (a), (c), (e) are from the dataset LSUI, while (b), (d), (f) are from the dataset UIEB. (a) (b) (c) (d) (e) (f) (g) [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: For each row, columns (a) to (g) represent the underwater image (a), the color corrected image (b), the dehazed image (c), the DINO feature of the color corrected image (d), the DINO feature of the dehazed image (e), the feature difference of (d) and (e), and the generated mask based on SAM2 from automatic prompt, respectively. Method UWCNN Replica PSNR SSIM PSNR SSIM SemiUIR[14] - - 17.3238 0.7902 PUGAN[… view at source ↗
Figure 13
Figure 13. Figure 13: Illustration of the intermediate results. For each subfigure, the first row shows the underwater image, the color corrected image, the dehazed image, and the final enhanced image, respectively; while the second row represents the corresponding semantic feature extracted from DINO model and visualized by PCA. Raw CCL HCLR WF-Diff CE-VAE SS-UIE DPF-Net PFUSIE UIE-CLIP Ours GT [PITH_FULL_IMAGE:figures/full_… view at source ↗
Figure 14
Figure 14. Figure 14: Quantitative performance comparison on the syn￾thetic dataset UWCNN [21]. 4.3. Ablation Study and Discussions To evaluate the effectiveness of the proposed design, three different ablation experiments were conducted. First, we directly fine-tuned a diffusion model using dataset UIEB for end-to-end UIE task, which means the heavily biased labels are utilized for supervised learning. The results is denoted … view at source ↗
Figure 15
Figure 15. Figure 15: Ablation study of partial training strategy. For each row, columns (a), (b), (c) are the underwater image, the enhanced image with/without partial training strategy, respectively. columns (d) and (e) are the zoom-in details from (b) and (c) to compare the cut-out effect on the edges. and testing due to effective segmentation and 2) those with useless segmentation results can only be used for testing, whic… view at source ↗
Figure 16
Figure 16. Figure 16: Performance analysis of Dino-based automatic prompt generation strategy. For each scene, subfigure (a)-(e) represent the color corrected image, the raw underwater image, the segmentation image, the feature difference between the color corrected image and haze removed image, and the final result, respectively. The blue and green points in (a) are the foreground/background prompts extracted based on (d) for… view at source ↗
Figure 18
Figure 18. Figure 18: Examples of the downstream applications about feature extraction (a), edge (b) and saliency detection (c). The number in the images of subfigure (a) denotes the amount of detected features. mobile underwater platforms equipped with computational graphics like Jetson Nano, and the leading performance of our proposal represents a reasonable trade-off for the com￾putational resources, particularly in applica… view at source ↗
Figure 19
Figure 19. Figure 19: Examples of the downstream applications about feature extraction (a), edge (b) and saliency detection (c). The number in the images of subfigure (a) denotes the amount of detected features. (a) (b) (c) (d) [PITH_FULL_IMAGE:figures/full_fig_p018_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: The result of feature matching by the raw under￾water images (a), (c), and the corresponding images enhanced by the proposal (b), (d). and SLAM. Future works will be dedicated to solving the above two issues. Despite that, the innovation of our pro￾posal that leveraging the multi-priors from relevant vision tasks to cross-domain supervision as well as physics-based task decomposition is validated and show… view at source ↗
read the original abstract

The underwater images are captured within diverse water-medium conditions, leading to complex degradation, including color bias, low contrast, and blur effect. Recently, learning-based methods have demonstrated their potential for underwater image enhancement (UIE). However, most of the previous work focus on the training strategy or network design to make the enhanced result aligned well with the labels in datasets, ignoring that the labels are selected from the enhanced results of previous UIE methods and these pseudo-labels are noisy. Consequently, the performance of their models is not satisfactory to a certain extent. However, collecting the true labels of the underwater images is challenging. In this work, we propose a transfer learning-based UIE that does not require underwater images to have paired noisy or true labels for learning. Instead, the UIE task is first divided into global color correction, haze removal, and background noise suppression following the underwater physics. Then multiple types of prior from other vision tasks are leveraged as cross-domain supervision in each step. In this way, a novel UIE is available via transfer learning, and the physics-aligned UIE decomposition provides theoretical soundness. Qualitative and quantitative experiments demonstrate that our proposal based on physics and priors fusion achieves SOTA performance in the UIE task and effectively boosts downstream vision tasks, significantly outperforming benchmark methods. Project repo: https://github.com/Haru2022/P2-UIE.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a transfer learning approach for underwater image enhancement (UIE) that decomposes the task into three steps—global color correction, haze removal, and background noise suppression—based on underwater physics, and uses priors transferred from other vision tasks as supervision for each step without requiring paired underwater labels. It claims this yields theoretically sound results and achieves SOTA performance on UIE and downstream tasks.

Significance. If the decomposition accurately models the physics and the priors transfer effectively, the method could provide a label-free alternative to supervised UIE methods that rely on noisy pseudo-labels, potentially improving generalization and performance in real underwater scenarios.

major comments (2)
  1. [Abstract] Abstract: The assertion that the decomposition into global color correction, haze removal, and background noise suppression accurately captures the dominant physical effects and provides theoretical soundness is not supported by explicit comparison to the standard Jaffe-McGlamery underwater image formation model, which couples color attenuation, scattering, and transmission in a single equation; this risks unmodeled cross-effects (e.g., color-dependent scattering) that independent cross-domain priors would not correct.
  2. [Method] Method description: The paper must show that the composition of the three modules reconstructs the full degradation operator; without this verification, the transfer-learning argument that priors can supervise each step independently collapses, undermining attribution of any SOTA gains to the physics-prior fusion.
minor comments (1)
  1. [Abstract] Abstract: The claim of SOTA performance via qualitative and quantitative experiments is stated without any numerical results, specific baselines, dataset names, or metrics, which reduces the ability to evaluate the central claim from the provided text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments regarding the theoretical grounding of our physics-based decomposition. We respond to each major comment below and will revise the manuscript accordingly to address the concerns.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The assertion that the decomposition into global color correction, haze removal, and background noise suppression accurately captures the dominant physical effects and provides theoretical soundness is not supported by explicit comparison to the standard Jaffe-McGlamery underwater image formation model, which couples color attenuation, scattering, and transmission in a single equation; this risks unmodeled cross-effects (e.g., color-dependent scattering) that independent cross-domain priors would not correct.

    Authors: We agree that the current abstract and manuscript lack an explicit side-by-side comparison to the Jaffe-McGlamery model. Our decomposition separates the dominant effects (wavelength-dependent absorption for color bias, scattering for haze, and additive noise) to enable independent cross-domain prior supervision, which is a practical approximation used in much of the UIE literature. To directly address the risk of unmodeled cross-effects, we will add a new subsection in the revised manuscript that compares our decomposition to the standard model, discusses potential interactions such as color-dependent scattering, and notes the approximation's limitations. revision: yes

  2. Referee: [Method] Method description: The paper must show that the composition of the three modules reconstructs the full degradation operator; without this verification, the transfer-learning argument that priors can supervise each step independently collapses, undermining attribution of any SOTA gains to the physics-prior fusion.

    Authors: The current manuscript motivates the three-module decomposition from underwater physics but does not include a formal verification that their composition exactly inverts the full degradation operator. The independent prior supervision is presented as valid because each module targets a separable physical component. We will revise the method section to include verification of the composition, for example by applying the modules to synthetically degraded images generated from a forward model and measuring reconstruction fidelity, thereby supporting the attribution of gains to the physics-prior approach. revision: yes

Circularity Check

0 steps flagged

No circularity: decomposition and priors are external to fitted data

full rationale

The paper asserts a physics-based decomposition of UIE into global color correction, haze removal, and background noise suppression, then applies cross-domain priors as supervision without paired underwater labels. No equations, derivations, or fitted parameters are shown that reduce any output to quantities defined from the same data or self-citations. The method is presented as relying on external priors and standard underwater physics references, with no load-bearing self-citation chains or self-definitional steps visible in the abstract or description. This is the common case of a self-contained proposal whose central claims rest on independent assumptions rather than internal redefinition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no equations or implementation details, so no free parameters, axioms, or invented entities can be identified.

pith-pipeline@v0.9.1-grok · 5792 in / 1049 out tokens · 46460 ms · 2026-06-27T03:58:55.298421+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 3 linked inside Pith

  1. [1]

    Dense-haze:A benchmarkforimagedehazingwithdense-hazeandhaze-freeimages, in: 2019 IEEE international conference on image processing (ICIP), IEEE

    Ancuti,C.O.,Ancuti,C.,Sbert,M.,Timofte,R.,2019. Dense-haze:A benchmarkforimagedehazingwithdense-hazeandhaze-freeimages, in: 2019 IEEE international conference on image processing (ICIP), IEEE. pp. 1014–1018

    2019

  2. [2]

    A computational approach to edge detection

    Canny, J., 2009. A computational approach to edge detection. IEEE Transactions on pattern analysis and machine intelligence , 679–698

    2009

  3. [3]

    Cao, J., Zeng, Z., Zhang, X., Zhang, H., Fan, C., Jiang, G., Lin, W.,

  4. [4]

    Pattern Recognition 162, 111395

    Unveilingtheunderwaterworld:Clipperceptionmodel-guided underwater image enhancement. Pattern Recognition 162, 111395

  5. [5]

    Underwater image enhancement by wavelengthcompensationanddehazing

    Chiang, J.Y., Chen, Y.C., 2011. Underwater image enhancement by wavelengthcompensationanddehazing. IEEEtransactionsonimage processing 21, 1756–1769

    2011

  6. [6]

    Pugan: Physical model-guided underwater image enhancement using gan with dual-discriminators

    Cong, R., Yang, W., Zhang, W., Li, C., Guo, C.L., Huang, Q., Kwong, S., 2023. Pugan: Physical model-guided underwater image enhancement using gan with dual-discriminators. IEEE Transactions on Image Processing 32, 4472–4485. First Author et al.:Preprint submitted to ElsevierPage 18 of 20 Short Title of the Article

    2023

  7. [7]

    Cong, Z., Zhou, Y., Wu, L., Tian, L., Chen, Z., Guan, M., He, L.,

  8. [8]

    InformationFusion,103732

    Pgf-net: Fusing a physical imaging model with self-attention forrobustunderwaterfeaturedetection. InformationFusion,103732

  9. [9]

    Underwater depth estimation and image restoration based on single images

    Drews,P.L.,Nascimento,E.R.,Botelho,S.S.,Campos,M.F.M.,2016. Underwater depth estimation and image restoration based on single images. IEEE computer graphics and applications 36, 24–35

    2016

  10. [10]

    Fine-tuning image-conditional diffusion models is easier than you think, in: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), IEEE

    Garcia, G.M., Abou Zeid, K., Schmidt, C., De Geus, D., Hermans, A., Leibe, B., 2025. Fine-tuning image-conditional diffusion models is easier than you think, in: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), IEEE. pp. 753–762

    2025

  11. [11]

    A survey on under- water computer vision

    González-Sabbagh, S.P., Robles-Kelly, A., 2023. A survey on under- water computer vision. ACM Computing Surveys 55, 1–39

    2023

  12. [12]

    Underwater ranker: Learn which is better and how to be better, in: Proceedings of the AAAI conference on artificial intelligence, pp

    Guo, C., Wu, R., Jin, X., Han, L., Zhang, W., Chai, Z., Li, C., 2023. Underwater ranker: Learn which is better and how to be better, in: Proceedings of the AAAI conference on artificial intelligence, pp. 702–709

    2023

  13. [13]

    Uw-gan: Single-image depth estimation and image enhancement for underwater images

    Hambarde, P., Murala, S., Dhall, A., 2021. Uw-gan: Single-image depth estimation and image enhancement for underwater images. IEEE Transactions on Instrumentation and Measurement 70, 1–12

    2021

  14. [14]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., Abbeel, P., 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33, 6840–6851

    2020

  15. [15]

    Underwater sequential images enhancement via diffusion and physics priors fusion

    Hu, H., Bin, Y., Wen, C.y., Wang, B., 2025. Underwater sequential images enhancement via diffusion and physics priors fusion. Infor- mation Fusion , 103365

    2025

  16. [16]

    Contrastivesemi- supervised learning for underwater image restoration via reliable bank, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Huang,S.,Wang,K.,Liu,H.,Chen,J.,Li,Y.,2023. Contrastivesemi- supervised learning for underwater image restoration via reliable bank, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 18145–18155

    2023

  17. [17]

    Fastunderwaterimageenhance- mentforimprovedvisualperception

    Islam,M.J.,Xia,Y.,Sattar,J.,2020. Fastunderwaterimageenhance- mentforimprovedvisualperception. IEEERoboticsandAutomation Letters 5, 3227–3234

    2020

  18. [18]

    Computer modeling and the design of optimal underwater imaging systems

    Jaffe, J.S., 1990. Computer modeling and the design of optimal underwater imaging systems. IEEE Journal of Oceanic Engineering 15, 101–111

    1990

  19. [19]

    Image enhancement via associated perturbation removal and texture reconstructionlearning

    Jiang, K., Wang, R., Xiao, Y., Jiang, J., Xu, X., Lu, T., 2024. Image enhancement via associated perturbation removal and texture reconstructionlearning. IEEE/CAAJournalofAutomaticaSinica11, 2253–2269

    2024

  20. [20]

    Jiang, K., Wang, Z., Wang, Z., Chen, C., Yi, P., Lu, T., Lin, C.W.,

  21. [21]

    1078–1086

    Degrade is upgrade: Learning degradation for low-light image enhancement, in: Proceedings of the AAAI conference on artificial intelligence, pp. 1078–1086

  22. [22]

    5148–5157

    Ke,J.,Wang,Q.,Wang,Y.,Milanfar,P.,Yang,F.,2021.Musiq:Multi- scale image quality transformer, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 5148–5157

    2021

  23. [23]

    Rad:Region-awarediffusionmodels for image inpainting, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp

    Kim,S.,Suh,S.,Lee,M.,2025. Rad:Region-awarediffusionmodels for image inpainting, in: Proceedings of the Computer Vision and Pattern Recognition Conference, pp. 2439–2448

    2025

  24. [24]

    Underwater scene prior inspired deep underwater image and video enhancement

    Li, C., Anwar, S., Porikli, F., 2020. Underwater scene prior inspired deep underwater image and video enhancement. Pattern Recognition 98, 107038

    2020

  25. [25]

    An underwater image enhancement benchmark dataset and beyond

    Li,C.,Guo,C.,Ren,W.,Cong,R.,Hou,J.,Kwong,S.,Tao,D.,2019. An underwater image enhancement benchmark dataset and beyond. IEEE transactions on image processing 29, 4376–4389

    2019

  26. [26]

    Underwater image captioning via attention mechanism based fusion of visual and textual information

    Li, L., Li, H., Ren, P., 2025a. Underwater image captioning via attention mechanism based fusion of visual and textual information. Information Fusion , 103269

  27. [27]

    S2g-gcn: A plot classi- fication network integrating spectrum-to-graph modeling and graph convolutional network for compact hfswr

    Li, X., Sun, W., Ji, Y., Huang, W., 2025b. S2g-gcn: A plot classi- fication network integrating spectrum-to-graph modeling and graph convolutional network for compact hfswr. IEEE Geoscience and Remote Sensing Letters

  28. [28]

    A visual-textual mutual guidance fusion network for remote sensing visual question answering

    Liu,H.,Chen,L.,Lu,X.,Wang,H.,Bai,L.,Wang,M.,Ren,P.,2026. A visual-textual mutual guidance fusion network for remote sensing visual question answering. Pattern Recognition 176, 113258

    2026

  29. [29]

    Underwater imageenhancementwithcascadedcontrastivelearning

    Liu, Y., Jiang, Q., Wang, X., Luo, T., Zhou, J., 2024. Underwater imageenhancementwithcascadedcontrastivelearning. IEEETrans- actions on Multimedia

    2024

  30. [30]

    From synthetic to real: Image dehazing collaborating with unlabeled real data, in: Proceedings of the 29th ACM international conference on multimedia, pp

    Liu, Y., Zhu, L., Pei, S., Fu, H., Qin, J., Zhang, Q., Wan, L., Feng, W., 2021. From synthetic to real: Image dehazing collaborating with unlabeled real data, in: Proceedings of the 29th ACM international conference on multimedia, pp. 50–58

    2021

  31. [31]

    Distinctive image features from scale-invariant keypoints

    Lowe, D.G., 2004. Distinctive image features from scale-invariant keypoints. International journal of computer vision 60, 91–110

    2004

  32. [32]

    Visual-instructed degradation diffusion for all-in-one imagerestoration,in:ProceedingsoftheComputerVisionandPattern Recognition Conference, pp

    Luo,W.,Qin,H.,Chen,Z.,Wang,L.,Zheng,D.,Li,Y.,Liu,Y.,Li,B., Hu, W., 2025. Visual-instructed degradation diffusion for all-in-one imagerestoration,in:ProceedingsoftheComputerVisionandPattern Recognition Conference, pp. 12764–12777

    2025

  33. [33]

    A computer model for underwater camera systems, in: Ocean Optics VI, SPIE

    McGlamery, B., 1980. A computer model for underwater camera systems, in: Ocean Optics VI, SPIE. pp. 221–231

    1980

  34. [34]

    Dpf-net: Physical imaging model embedded data-driven underwater image enhance- ment

    Mei, H., Li, K., Liu, S., Ma, C., Jiang, Q., 2025. Dpf-net: Physical imaging model embedded data-driven underwater image enhance- ment. ISPRS Journal of Photogrammetry and Remote Sensing 228, 679–693

    2025

  35. [35]

    Bi-level inter- modality modulation for unsupervised visible-infrared person re- identification

    Peng, J., Liu, J., Zuo, X., Tao, Z., Wang, H., 2026. Bi-level inter- modality modulation for unsupervised visible-infrared person re- identification. IEEE Transactions on Information Forensics and Security

    2026

  36. [36]

    Adaptive dual-domain learning for under- water image enhancement, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Peng, L., Bian, L., 2025. Adaptive dual-domain learning for under- water image enhancement, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 6461–6469

    2025

  37. [37]

    U-shapetransformerforunderwater image enhancement

    Peng,L.,Zhu,C.,Bian,L.,2023. U-shapetransformerforunderwater image enhancement. IEEE Transactions on Image Processing 32, 3066–3079

    2023

  38. [38]

    Ce-vae: Capsule enhanced variational autoencoderforunderwaterimageenhancement,in:2025IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), IEEE

    Pucci, R., Martinel, N., 2025. Ce-vae: Capsule enhanced variational autoencoderforunderwaterimageenhancement,in:2025IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), IEEE. pp. 2113–2123

    2025

  39. [39]

    Underwaterimageco-enhancementwithcorrelationfeaturematching and joint learning

    Qi,Q.,Zhang,Y.,Tian,F.,Wu,Q.J.,Li,K.,Luan,X.,Song,D.,2021. Underwaterimageco-enhancementwithcorrelationfeaturematching and joint learning. IEEE Transactions on Circuits and Systems for Video Technology 32, 1133–1147

    2021

  40. [40]

    Sam 2: Segment anything in images and videos

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al., 2024. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714

    Pith/arXiv arXiv 2024

  41. [41]

    High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Rombach,R.,Blattmann,A.,Lorenz,D.,Esser,P.,Ommer,B.,2022. High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10684–10695

    2022

  42. [42]

    Recoveryof underwatervisibility and structure by polarization analysis

    Schechner,Y.Y., Karpel,N.,2005. Recoveryof underwatervisibility and structure by polarization analysis. IEEE Journal of oceanic engineering 30, 570–587

    2005

  43. [43]

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khalidov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., et al.,

  44. [44]

    arXiv preprint arXiv:2508.10104

    Dinov3. arXiv preprint arXiv:2508.10104

    Pith/arXiv arXiv

  45. [45]

    The replica dataset: A digital replica of indoor spaces

    Straub, J., Whelan, T., Ma, L., Chen, Y., Wijmans, E., Green, S., Engel, J.J., Mur-Artal, R., Ren, C., Verma, S., et al., 2019. The replica dataset: A digital replica of indoor spaces. arXiv preprint arXiv:1906.05797

    Pith/arXiv arXiv 2019

  46. [46]

    Varghese, N., Kumar, A., Rajagopalan, A., 2023. Self-supervised monocular underwater depth recovery, image restoration, and a real- sea video dataset, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 12248–12258

    2023

  47. [47]

    Tensorized parameter-free multi-view spectral clustering based on fair representation learning

    Wang, H., Chen, Y., Yao, M., Liu, Q., Jiang, G., Fu, X., 2026a. Tensorized parameter-free multi-view spectral clustering based on fair representation learning. IEEE Transactions on Multimedia

  48. [48]

    Tensorized fine-grained incompletemultiviewclusteringviaintrinsicstructurerecovery.IEEE Transactions on Systems, Man, and Cybernetics: Systems

    Wang, H., Cui, L., Yao, M., Fu, X., 2026b. Tensorized fine-grained incompletemultiviewclusteringviaintrinsicstructurerecovery.IEEE Transactions on Systems, Man, and Cybernetics: Systems

  49. [49]

    Watercycledif- fusion: Visual-textual fusion empowered underwater image enhance- ment

    Wang, H., Zhang, W., Xu, Y., Li, H., Ren, P., 2025. Watercycledif- fusion: Visual-textual fusion empowered underwater image enhance- ment. Information Fusion , 103693

    2025

  50. [50]

    Tartanair: A dataset to push the limitsofvisualslam,in:2020IEEE/RSJInternationalConferenceon Intelligent Robots and Systems (IROS), IEEE

    Wang, W., Zhu, D., Wang, X., Hu, Y., Qiu, Y., Wang, C., Hu, Y., Kapoor, A., Scherer, S., 2020. Tartanair: A dataset to push the limitsofvisualslam,in:2020IEEE/RSJInternationalConferenceon Intelligent Robots and Systems (IROS), IEEE. pp. 4909–4916. First Author et al.:Preprint submitted to ElsevierPage 19 of 20 Short Title of the Article

    2020

  51. [51]

    Multi-axis feature diversity enhancement for remote sensing video super-resolution

    Xiao, Y., Yuan, Q., Jiang, K., Chen, Y., Wang, S., Lin, C.W., 2025. Multi-axis feature diversity enhancement for remote sensing video super-resolution. IEEE Transactions on Image Processing

    2025

  52. [52]

    Ediffsr: An efficient diffusion probabilistic model for remote sensing image super-resolution

    Xiao,Y.,Yuan,Q.,Jiang,K.,He,J.,Jin,X.,Zhang,L.,2023. Ediffsr: An efficient diffusion probabilistic model for remote sensing image super-resolution. IEEE Transactions on Geoscience and Remote Sensing 62, 1–14

    2023

  53. [53]

    Xie, Y., Kong, L., Chen, K., Zheng, Z., Yu, X., Yu, Z., Zheng, B., 2024.Uveb:Alarge-scalebenchmarkandbaselinetowardsreal-world underwater video enhancement, in: Proceedings of the IEEE/CVF ConferenceonComputerVisionandPatternRecognition,pp.22358– 22367

    2024

  54. [54]

    Saliency detection via graph-based manifold ranking, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Yang, C., Zhang, L., Lu, H., Ruan, X., Yang, M.H., 2013. Saliency detection via graph-based manifold ranking, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3166–3173

    2013

  55. [55]

    Ying,Z.,Niu,H.,Gupta,P.,Mahajan,D.,Ghadiyaram,D.,Bovik,A.,

  56. [56]

    3575–3585

    From patches to pictures (paq-2-piq): Mapping the perceptual spaceofpicturequality,in:ProceedingsoftheIEEE/CVFconference on computer vision and pattern recognition, pp. 3575–3585

  57. [57]

    Underwater image enhancement via minimal color loss and locally adaptive contrast enhancement

    Zhang, W., Zhuang, P., Sun, H.H., Li, G., Kwong, S., Li, C., 2022. Underwater image enhancement via minimal color loss and locally adaptive contrast enhancement. IEEE Transactions on Image Pro- cessing 31, 3997–4010

    2022

  58. [58]

    Zhao, C., Cai, W., Dong, C., Hu, C., 2024. Wavelet-based fourier in- formation interaction with frequency diffusion adjustment for under- waterimagerestoration,in:ProceedingsoftheIEEE/CVFConference on Computer Vision and Pattern Recognition, pp. 8281–8291

    2024

  59. [59]

    A novel underwater target tracking method in uasns via collaborative deep reinforcement learning

    Zheng, L., Liu, M., Zhang, S., Dong, S., 2025. A novel underwater target tracking method in uasns via collaborative deep reinforcement learning. Information Fusion , 103797

    2025

  60. [60]

    International Journal of Computer Vision 132, 4132–4156

    Zhou, J., Sun, J., Li, C., Jiang, Q., Zhou, M., Lam, K.M., Zhang, W., Fu,X.,2024.Hclr-net:hybridcontrastivelearningregularizationwith locally randomized perturbation for underwater image enhancement. International Journal of Computer Vision 132, 4132–4156

    2024

  61. [61]

    Underwater image enhancement with hyper-laplacian reflectance priors

    Zhuang, P., Wu, J., Porikli, F., Li, C., 2022. Underwater image enhancement with hyper-laplacian reflectance priors. IEEE Trans- actions on Image Processing 31, 5442–5455. First Author et al.:Preprint submitted to ElsevierPage 20 of 20

    2022