pith. sign in

arxiv: 2605.14677 · v1 · pith:C3LRI25Knew · submitted 2026-05-14 · 📡 eess.IV

An Attention-Enhanced Network with Joint Dehazing and Retinex-Based Enhancement for Underwater Images

Sahana Ray , Bibhabasu Debnath , Sanjay Ghosh This is my paper

Pith reviewed 2026-06-30 20:03 UTC · model grok-4.3

classification 📡 eess.IV
keywords underwater image enhancementdehazingRetinexattention mechanismU-Net++image restorationimage formation model
0
0 comments X

The pith

A three-stage network extends the underwater image formation model with Retinex enhancement and attention refinement to restore degraded images competitively.

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

The paper introduces the ADR network to address wavelength-dependent absorption, scattering, and turbidity in underwater images that impair applications like vehicle navigation and marine inspection. It builds a hybrid pipeline that first applies an extended image formation model for dehazing, then uses Retinex principles for further enhancement, and finally refines the output with an attention-augmented U-Net++. The approach seeks to retain physical grounding while gaining from learned components, and tests show it performs on par with leading methods on standard benchmarks. A sympathetic reader would value this if the combined stages produce clearer, more usable images than either pure physical models or black-box networks alone.

Core claim

The ADR network performs underwater image restoration through three sequential stages: dehazing via an extended image formation model with added terms, Retinex-based contrast and color enhancement, and final refinement by an attention-enabled U-Net++ architecture, achieving competitive quantitative and qualitative results against state-of-the-art methods on the UIEB and UFO-120 datasets.

What carries the argument

The three-stage ADR pipeline that couples an extended physical image formation model for dehazing with Retinex enhancement and attention-driven U-Net++ refinement.

If this is right

  • The extended formation model supplies an interpretable starting point that pure data-driven networks lack.
  • Retinex enhancement after dehazing improves local contrast and color balance in the restored images.
  • Attention mechanisms in the final U-Net++ stage allow selective focus on remaining artifacts without global over-correction.
  • Competitive benchmark scores indicate the hybrid method can serve as a drop-in replacement for existing underwater restoration pipelines in AUV and inspection workflows.

Where Pith is reading between the lines

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

  • The same staged structure could be tested on other scattering media such as fog or biomedical tissue images where physical models are partially known.
  • Replacing the final attention U-Net++ with a lighter refinement block might reduce inference time for real-time vehicle use while preserving most gains.
  • Measuring downstream task performance, such as object detection accuracy on restored versus raw frames, would show whether the visual improvements translate to practical utility.

Load-bearing premise

Adding extra terms to the classical image formation model will capture enough of the nonlinear wavelength-dependent absorption and scattering to make the subsequent stages effective.

What would settle it

A direct comparison on the UIEB dataset showing that the full ADR pipeline yields lower PSNR or SSIM scores than either the dehazing stage alone or a standard U-Net++ baseline would undermine the value of the joint extension and multi-stage design.

read the original abstract

Underwater images suffer from severe wavelength-dependent light absorption and scattering, and turbidity due to suspended particles, degrading visual quality for applications in autonomous underwater vehicles (AUVs), marine biology, archaeology, and offshore infrastructure inspection. Classical IFM inadequately capture nonlinear underwater light behavior, while purely data-driven methods lack physical interpretability. This paper proposes a three-stage network named ADR, that extends the underwater image formation model with additional terms to perform underwater dehazing, followed by Retinex-based enhancement and attention-enabled U-Net++ refinement. Experiments on UIEB and UFO-120 benchmark datasets demonstrate competitive performance with state-of-the-art methods.

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 three-stage ADR network for underwater image enhancement. Stage 1 extends the classical underwater image formation model (IFM) with additional terms to perform dehazing; stage 2 applies Retinex-based enhancement; stage 3 uses an attention-enabled U-Net++ for refinement. Experiments on the UIEB and UFO-120 benchmarks are reported to demonstrate competitive performance against state-of-the-art methods.

Significance. A hybrid physical-plus-data-driven pipeline that retains interpretability while addressing the known limitations of classical IFM could be valuable for AUV navigation and marine inspection if the extension is shown to capture the missing nonlinear wavelength dependence and if the performance claims are supported by ablations and statistical controls.

major comments (2)
  1. [Abstract] Abstract: the central modeling claim—that extending the classical IFM with unspecified additional terms remedies its failure to capture nonlinear wavelength-dependent absorption and scattering—is presented without any functional form, derivation, or comparison to measured attenuation spectra. This assumption is load-bearing for both the physical-interpretability argument and the assertion that the three-stage pipeline outperforms purely data-driven baselines.
  2. [Experiments] Experiments (UIEB and UFO-120 results): no training protocol, hyper-parameter settings, ablation studies, or error bars are supplied. Without these, the reported competitive performance cannot be attributed to the proposed IFM extension versus the Retinex or attention modules, undermining the cross-method comparison.
minor comments (1)
  1. Notation for the extended IFM terms should be introduced with explicit variable definitions and physical units once the equations are added.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and will incorporate the suggested changes in the revised version.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central modeling claim—that extending the classical IFM with unspecified additional terms remedies its failure to capture nonlinear wavelength-dependent absorption and scattering—is presented without any functional form, derivation, or comparison to measured attenuation spectra. This assumption is load-bearing for both the physical-interpretability argument and the assertion that the three-stage pipeline outperforms purely data-driven baselines.

    Authors: We agree that the abstract does not explicitly state the functional form or derivation of the IFM extension. The full manuscript details the extended model in Section 3, but we acknowledge the abstract requires clarification to support the interpretability claims. In the revision we will update the abstract to specify the additional terms, include a concise derivation addressing nonlinear wavelength dependence, and reference a comparison against measured attenuation spectra. revision: yes

  2. Referee: [Experiments] Experiments (UIEB and UFO-120 results): no training protocol, hyper-parameter settings, ablation studies, or error bars are supplied. Without these, the reported competitive performance cannot be attributed to the proposed IFM extension versus the Retinex or attention modules, undermining the cross-method comparison.

    Authors: We agree that the experimental section lacks these details. We will revise the manuscript to add a full description of the training protocol and hyper-parameters, comprehensive ablation studies isolating each component (IFM extension, Retinex, and attention U-Net++), and error bars or standard deviations from repeated runs to enable proper attribution of performance gains. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation chain not present in inspectable text

full rationale

The provided abstract and context contain no equations, functional forms, fitting procedures, or derivation steps. The central claim is a proposal of a three-stage ADR network that extends the classical IFM, but without any quoted mathematical reduction or self-citation chain that collapses to its own inputs, no circularity of any enumerated kind can be exhibited. The modeling extension is asserted as a contribution rather than shown to be equivalent to fitted data or prior self-citations by construction. This is the expected honest non-finding when the source text supplies no load-bearing derivation to analyze.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted.

pith-pipeline@v0.9.1-grok · 5637 in / 1097 out tokens · 24176 ms · 2026-06-30T20:03:49.384580+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

32 extracted references · 4 canonical work pages · 2 internal anchors

  1. [1]

    Long wave- lengths attenuate rapidly, causing color distortion and low contrast, while scattering introduces haze

    INTRODUCTION Light propagation underwater differs from atmospheric con- ditions due to water’s optical properties [1, 2]. Long wave- lengths attenuate rapidly, causing color distortion and low contrast, while scattering introduces haze. Still, underwater images remain crucial for ecological and biological research, including species monitoring and environ...

  2. [2]

    An Attention-Enhanced Network with Joint Dehazing and Retinex-Based Enhancement for Underwater Images

    PROPOSED METHOD The proposed ADR framework employs a three-stage archi- tecture that addresses the specific degradation challenges en- countered in underwater imaging, as illustrated in Figure 1. arXiv:2605.14677v1 [eess.IV] 14 May 2026 2.1. Enhanced Physics-Guided Dehazing The first stage implements an extended physics-based correc- tion model. To accoun...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  3. [3]

    EXPERIMENTS 3.1. Dataset and Implementation Details The proposed model was trained and evaluated on the UIEB (Underwater Image Enhancement Benchmark) dataset [17], a widely adopted benchmark containing 890 paired under- water images, partitioned into 700 training and 190 test im- ages. Additionally, the model was trained and evaluated on the UFO-120 [22] ...

    2016

  4. [4]

    [11] [17] [20] [18] [19] ADR Time(s) 0.32 38.71 0.610.090.45 0.33 0.53 Table 4: Ablation study on UIEB test set (256×256). Model PSNR↑SSIM↑LPIPS↓ Full model22.970.90400.1175 w/o Turbidity term 22.90 0.8997 0.1183 w/o Noise term23.110.9026 0.1168 w/o Retinex stage 22.97 0.9031 0.1159 w/o U-Net++ stage 22.85 0.90330.1146 w/oL perc 22.87 0.9026 0.1191 w/oL S...

    work page arXiv

  5. [5]

    Experiments on the UIEB and UFO datasets demonstrate that the proposed method achieves competitive performance in terms of PSNR, SSIM, and LPIPS

    CONCLUSION This work proposed a three-stage UIE network that combines physics-based modeling and deep learning. Experiments on the UIEB and UFO datasets demonstrate that the proposed method achieves competitive performance in terms of PSNR, SSIM, and LPIPS. By improving visibility, color fidelity, and structural clarity, this work offers a practical solut...

  6. [6]

    Light in the sea,

    Seibert Q Duntley, “Light in the sea,”Journal of the Optical Society of America, vol. 53, no. 2, pp. 214–233, 1963

    1963

  7. [7]

    Recovery of under- water visibility and structure by polarization analysis,

    Yoav Y Schechner and Nir Karpel, “Recovery of under- water visibility and structure by polarization analysis,” IEEE Journal of Oceanic Engineering, vol. 30, no. 3, pp. 570–587, 2006

    2006

  8. [8]

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

    Weidong Zhang, Peixian Zhuang, Hai-Han Sun, Guo- hou Li, Sam Kwong, and Chongyi Li, “Underwater image enhancement via minimal color loss and locally adaptive contrast enhancement,”IEEE Transactions on Image Processing, vol. 31, pp. 3997–4010, 2022

    2022

  9. [9]

    Underwater image enhance- ment via piecewise color correction and dual prior opti- mized contrast enhancement,

    Weidong Zhang, Songlin Jin, Peixian Zhuang, Zheng Liang, and Chongyi Li, “Underwater image enhance- ment via piecewise color correction and dual prior opti- mized contrast enhancement,”IEEE Signal Processing Letters, vol. 30, pp. 229–233, 2023

    2023

  10. [10]

    Computer modeling and the design of optimal underwater imaging systems,

    Jules S Jaffe, “Computer modeling and the design of optimal underwater imaging systems,”IEEE Journal of Oceanic Engineering, vol. 15, no. 2, pp. 101–111, 2002

    2002

  11. [11]

    Chro- matic framework for vision in bad weather,

    Srinivasa G Narasimhan and Shree K Nayar, “Chro- matic framework for vision in bad weather,”Proc. IEEE Conference on Computer Vision and Pattern Recogni- tion, pp. 598–605, 2000

    2000

  12. [12]

    Single image dehazing,

    Raanan Fattal, “Single image dehazing,”ACM Trans- actions on Graphics, vol. 27, no. 3, pp. 1–9, 2008

    2008

  13. [13]

    Single image haze removal using dark channel prior,

    Kaiming He, Jian Sun, and Xiaoou Tang, “Single image haze removal using dark channel prior,”IEEE Transac- tions on Pattern Analysis and Machine Intelligence, vol. 33, no. 12, pp. 2341–2353, 2010

    2010

  14. [14]

    Transmission estimation in underwater single images,

    Paul Drews, Erickson Nascimento, Filipe Moraes, Silvia Botelho, and Mario Campos, “Transmission estimation in underwater single images,”Proc. IEEE International Conference on Computer Vision Workshops, pp. 825– 830, 2013

    2013

  15. [15]

    A revised under- water image formation model,

    Derya Akkaynak and Tali Treibitz, “A revised under- water image formation model,”Proc. IEEE Conference on Computer Vision and Pattern Recognition, pp. 6723– 6732, 2018

    2018

  16. [16]

    Underwater image restoration based on image blurriness and light absorption,

    Yan-Tsung Peng and Pamela C Cosman, “Underwater image restoration based on image blurriness and light absorption,”IEEE Transactions on Image Processing, vol. 26, no. 4, pp. 1579–1594, 2017

    2017

  17. [17]

    Lightness and retinex theory,

    Edwin H Land and John J McCann, “Lightness and retinex theory,”Journal of the Optical Society of Amer- ica, vol. 61, no. 1, pp. 1–11, 1971

    1971

  18. [18]

    Fast scale-adaptive bilateral texture smoothing,

    Sanjay Ghosh, Ruturaj G Gavaskar, Debasisha Panda, and Kunal N Chaudhury, “Fast scale-adaptive bilateral texture smoothing,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 30, no. 7, pp. 2015– 2026, 2019

    2015

  19. [19]

    Fast bright- pass bilateral filtering for low-light enhancement,

    Sanjay Ghosh and Kunal N Chaudhury, “Fast bright- pass bilateral filtering for low-light enhancement,”Proc. IEEE International Conference on Image Processing (ICIP), pp. 205–209, 2019

    2019

  20. [20]

    Un- derwater image enhancement via extended multi-scale retinex,

    Shu Zhang, Ting Wang, Junyu Dong, and Hui Yu, “Un- derwater image enhancement via extended multi-scale retinex,”Neurocomputing, vol. 245, pp. 1–9, 2017

    2017

  21. [21]

    On the duality between retinex and image dehazing,

    Adrian Galdran, Aitor Alvarez-Gila, Alessandro Bria, Javier Vazquez-Corral, and Marcelo Bertalm´ıo, “On the duality between retinex and image dehazing,”Proc. IEEE Conference on Computer Vision and Pattern Recognition, pp. 8212–8221, 2018

    2018

  22. [22]

    An un- derwater image enhancement benchmark dataset and be- yond,

    Chongyi Li, Chunle Guo, Wenqi Ren, Runmin Cong, Junhui Hou, Sam Kwong, and Dacheng Tao, “An un- derwater image enhancement benchmark dataset and be- yond,”IEEE Transactions on Image Processing, vol. 29, pp. 4376–4389, 2019

    2019

  23. [23]

    Underwater image enhancement via learning water type desensitized representations,

    Zhenqi Fu, Xiaopeng Lin, Wu Wang, Yue Huang, and Xinghao Ding, “Underwater image enhancement via learning water type desensitized representations,”Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2764–2768, 2022

    2022

  24. [24]

    An efficient approach for underwater image improvement: Deblurring, dehazing, and color correction,

    Alejandro Rico Espinosa, Declan McIntosh, and Alexandra Branzan Albu, “An efficient approach for underwater image improvement: Deblurring, dehazing, and color correction,”Proc. IEEE/CVF Winter Confer- ence on Applications of Computer Vision, pp. 206–215, 2023

    2023

  25. [25]

    Underwater image enhancement based on deep learning and image formation model,

    Xuelei Chen, Pin Zhang, Lingwei Quan, Chao Yi, and Cunyue Lu, “Underwater image enhancement based on deep learning and image formation model,”arXiv preprint arXiv:2101.00991, 2021

    work page arXiv 2021

  26. [26]

    UNet++: A Nested U-Net Architecture for Medical Image Segmentation

    Z Zhou, MMR Siddiquee, N Tajbakhsh, and J UNet+ Liang, “A nested U-Net architecture for medical image segmentation,”arXiv preprint arXiv:1807.10165, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  27. [27]

    Semantic segmen- tation of underwater imagery: Dataset and benchmark,

    Md Jahidul Islam, Chelsey Edge, Yuyang Xiao, Peigen Luo, Muntaqim Mehtaz, Christopher Morse, Sad- man Sakib Enan, and Junaed Sattar, “Semantic segmen- tation of underwater imagery: Dataset and benchmark,” pp. 1769–1776, 2020

    2020

  28. [28]

    The unreasonable ef- fectiveness of deep features as a perceptual metric,

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang, “The unreasonable ef- fectiveness of deep features as a perceptual metric,” Proc. IEEE Conference on Computer Vision and Pattern Recognition, pp. 586–595, 2018

    2018

  29. [29]

    Underwater image en- hancement via medium transmission-guided multi-color space embedding,

    Chongyi Li, Saeed Anwar, Junhui Hou, Runmin Cong, Chunle Guo, and Wenqi Ren, “Underwater image en- hancement via medium transmission-guided multi-color space embedding,”IEEE Transactions on Image Pro- cessing, vol. 30, pp. 4985–5000, 2021

    2021

  30. [30]

    Shallow-UWnet: Compressed model for underwater image enhancement (student abstract),

    Ankita Naik, Apurva Swarnakar, and Kartik Mittal, “Shallow-UWnet: Compressed model for underwater image enhancement (student abstract),” vol. 35, no. 18, pp. 15853–15854, 2021

    2021

  31. [31]

    Underwater Ranker: Learn which is better and how to be better,

    Chunle Guo, Ruiqi Wu, Xin Jin, Linghao Han, Weidong Zhang, Zhi Chai, and Chongyi Li, “Underwater Ranker: Learn which is better and how to be better,” inProceed- ings of the AAAI conference on artificial intelligence, 2023, vol. 37, pp. 702–709

    2023

  32. [32]

    Contrastive semi-supervised learning for underwater image restoration via reliable bank,

    Shirui Huang, Keyan Wang, Huan Liu, Jun Chen, and Yunsong Li, “Contrastive semi-supervised learning for underwater image restoration via reliable bank,” pp. 18145–18155, 2023

    2023