pith. sign in

arxiv: 1907.07863 · v1 · pith:U7PXWRDMnew · submitted 2019-07-17 · 💻 cs.CV · cs.LG· eess.IV

Diving Deeper into Underwater Image Enhancement: A Survey

Saeed Anwar , Chongyi Li This is my paper

Pith reviewed 2026-05-24 20:42 UTC · model grok-4.3

classification 💻 cs.CV cs.LGeess.IV
keywords underwater image enhancementdeep learningsurveybenchmarkimage formation modelevaluation metricsopen issues
0
0 comments X

The pith

This survey reviews deep learning methods for underwater image enhancement and benchmarks them on diverse datasets.

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

The paper consolidates the growing body of work on deep networks for underwater image enhancement into one reference. It covers image formation models that explain degradation, then details network architectures, training data, loss functions, and configurations used by existing methods. A systematic set of experiments compares the algorithms qualitatively and quantitatively across multiple datasets to create a benchmark. The survey also flags weaknesses in current datasets and metrics and lists open problems. A reader would value this because scattered papers make it hard to judge which approaches are reliable or where the field should go next.

Core claim

The paper establishes a reference by first presenting underwater image formation models, then surveying deep enhancement networks with attention to their architecture, parameters, training data, loss functions and configurations, summarizing datasets and metrics, running controlled comparisons of the algorithms, identifying shortcomings in existing benchmarks, and outlining unsolved open issues together with suggested research directions.

What carries the argument

The two-fold structure of a comprehensive review of deep networks plus a controlled experimental comparison that serves as a benchmark for the field.

If this is right

  • Researchers can consult the summarized network details and training setups when designing new models.
  • The identified shortcomings in datasets and metrics indicate concrete targets for improvement.
  • The listed open issues provide explicit directions that subsequent papers can address.
  • The benchmark comparison supplies a practical starting point for selecting methods on new underwater data.

Where Pith is reading between the lines

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

  • A shared open benchmark built from the surveyed datasets could reduce duplicated experimental effort across labs.
  • The image-formation models reviewed here may transfer directly to related degradation problems such as haze or low-light enhancement.
  • Releasing code and exact training splits for the benchmark comparisons would allow the community to verify and extend the results.

Load-bearing premise

The chosen deep algorithms, datasets and evaluation protocols are representative of the field and the comparison contains no undisclosed biases in implementation or selection.

What would settle it

A re-run of the same algorithms on the same datasets that produces substantially different performance orderings or conclusions about which methods are strongest would undermine the benchmark.

read the original abstract

The powerful representation capacity of deep learning has made it inevitable for the underwater image enhancement community to employ its potential. The exploration of deep underwater image enhancement networks is increasing over time, and hence; a comprehensive survey is the need of the hour. In this paper, our main aim is two-fold, 1): to provide a comprehensive and in-depth survey of the deep learning-based underwater image enhancement, which covers various perspectives ranging from algorithms to open issues, and 2): to conduct a qualitative and quantitative comparison of the deep algorithms on diverse datasets to serve as a benchmark, which has been barely explored before. To be specific, we first introduce the underwater image formation models, which are the base of training data synthesis and design of deep networks, and also helpful for understanding the process of underwater image degradation. Then, we review deep underwater image enhancement algorithms, and a glimpse of some of the aspects of the current networks is presented including network architecture, network parameters, training data, loss function, and training configurations. We also summarize the evaluation metrics and underwater image datasets. Following that, a systematically experimental comparison is carried out to analyze the robustness and effectiveness of deep algorithms. Meanwhile, we point out the shortcomings of current benchmark datasets and evaluation metrics. Finally, we discuss several unsolved open issues and suggest possible research directions. We hope that all efforts done in this paper might serve as a comprehensive reference for future research and call for the development of deep learning-based underwater image enhancement.

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 / 2 minor

Summary. The manuscript surveys deep learning-based underwater image enhancement, first reviewing underwater image formation models, then covering network architectures, parameters, training data, loss functions, and configurations of existing methods. It summarizes evaluation metrics and datasets, performs a systematic qualitative and quantitative comparison of selected deep algorithms across diverse datasets to establish a benchmark, identifies shortcomings in current datasets and metrics, and discusses open issues with suggested research directions.

Significance. If the benchmark comparison is representative and free of undisclosed selection or implementation biases, the work would provide a consolidated reference for the underwater image enhancement community, filling a gap by offering both an in-depth review and the first systematic experimental benchmark of deep methods, which could help standardize evaluation and highlight directions for future work.

major comments (2)
  1. [experimental comparison section (following the review of algorithms and metrics)] The experimental comparison section does not state explicit inclusion criteria for selecting the deep algorithms, datasets, or evaluation protocols used in the benchmark. This is load-bearing for the central claim that the comparison 'serves as a benchmark' because representativeness cannot be assessed without knowing the selection process or whether the chosen methods cover the range of architectures, losses, and training regimes reviewed earlier in the survey.
  2. [section describing the systematically experimental comparison] It is not specified whether the quantitative results were obtained by re-implementing all reviewed networks under identical hyperparameters and training configurations or by using author-provided code with potentially varying setups. This directly affects the validity of the robustness and effectiveness claims in the benchmark, as inconsistent implementation details could introduce biases not disclosed in the protocol.
minor comments (2)
  1. [abstract] The abstract contains a minor grammatical issue ('hence; a comprehensive survey') that should be corrected for clarity.
  2. [figures and tables in the experimental section] Some figure captions or table descriptions could more explicitly link back to the specific algorithms and datasets discussed in the review sections to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments on the experimental comparison section below and will revise the manuscript to provide the requested details.

read point-by-point responses

  1. Referee: The experimental comparison section does not state explicit inclusion criteria for selecting the deep algorithms, datasets, or evaluation protocols used in the benchmark. This is load-bearing for the central claim that the comparison 'serves as a benchmark' because representativeness cannot be assessed without knowing the selection process or whether the chosen methods cover the range of architectures, losses, and training regimes reviewed earlier in the survey.

    Authors: We agree that explicit inclusion criteria are not stated in the current manuscript. In the revision we will insert a dedicated paragraph (or subsection) immediately preceding the experimental results that specifies: (i) algorithm selection was limited to methods with publicly released code or sufficiently detailed architectures/losses to permit faithful re-implementation, chosen to span the main families reviewed earlier (CNN encoder-decoder, GAN-based, attention-augmented, etc.); (ii) datasets comprise the most widely adopted benchmarks containing both synthetic and real underwater images; (iii) evaluation protocols follow the standard metrics and train/test splits reported in the original papers. This addition will allow readers to judge representativeness directly. revision: yes

  2. Referee: It is not specified whether the quantitative results were obtained by re-implementing all reviewed networks under identical hyperparameters and training configurations or by using author-provided code with potentially varying setups. This directly affects the validity of the robustness and effectiveness claims in the benchmark, as inconsistent implementation details could introduce biases not disclosed in the protocol.

    Authors: We acknowledge the omission. The revised manuscript will explicitly describe the protocol: author-provided implementations were used when available and run with their original hyper-parameters; remaining networks were re-implemented from the papers and trained under a common set of settings (optimizer, learning-rate schedule, batch size, and hardware) chosen to be as close as possible to the originals while ensuring comparability. A supplementary table will list per-method training configurations and any necessary adaptations. This clarification will be added to the experimental section. revision: yes

Circularity Check

0 steps flagged

No circularity: survey and benchmark without derivation chain

full rationale

This is a survey paper whose central claims are coverage of existing algorithms, datasets, metrics, and a comparative benchmark experiment. No mathematical derivations, first-principles predictions, or fitted parameters are presented that could reduce to their own inputs by construction. The two-fold aim (comprehensive review plus experimental comparison) rests on selection and implementation choices rather than self-definitional equations or self-citation chains that substitute for independent evidence. The paper therefore contains no load-bearing steps matching any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

As a survey paper, no new free parameters, axioms, or invented entities are introduced; the work relies on existing literature and standard evaluation practices.

pith-pipeline@v0.9.0 · 5794 in / 1012 out tokens · 14967 ms · 2026-05-24T20:42:32.973742+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

65 extracted references · 65 canonical work pages · 9 internal anchors

  1. [1]

    In: CVPR (2018)

    Akkaynak, D., Treibitz, T.: A revised underwater image formation model. In: CVPR (2018)

    work page 2018

  2. [2]

    Densely Residual Laplacian Super-Resolution

    Anwar, S., Barnes, N.: Densely residual laplacian super- resolution. arXiv preprint arXiv:1906.12021 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1906

  3. [3]

    Deep Underwater Image Enhancement

    Anwar, S., Li, C., Porikli, F.: Deep underwater image enhancement. arXiv preprint arXiv:1807.03528 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    In: CVPR (2012)

    Aucuti, C., Ancuti, C.O., Bekaert, P.: Enhancing under- water images and videos by fusion. In: CVPR (2012)

    work page 2012

  5. [5]

    TPAMI (2017)

    Badrinarayanan, V., Kendall, A., Cipolla, R.: Segnet: A deep convolutional encoder-decoder architecture for im- age segmentation. TPAMI (2017)

    work page 2017

  6. [6]

    Underwater Single Image Color Restoration Using Haze-Lines and a New Quantitative Dataset

    Berman, D., Levy, D., Avidan, S., Treibitz, T.: Un- derwater single image color restoration using haze- lines and a new quantitative dataset. arXiv preprint arXiv:1811.01343 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  7. [7]

    In: Eco- logical Informatics (2014)

    Boom, B.J., He, J., Palazzo, S., Huang, P.X., Chou, H.M., Lin, F.P., Spampinato, C., Fisher, R.B.: A research tool for long-term and continuous analysis of fish assemblage in coral-reefs using underwater camera footage. In: Eco- logical Informatics (2014)

    work page 2014

  8. [8]

    In: SSIAI (2018)

    Cao, K., Peng, Y.T., Cosman, P.C.: Underwater image restoration using deep networks to estimate background light and scene depth. In: SSIAI (2018)

    work page 2018

  9. [9]

    IEEE Transac- tions on Image Processing 21(4), 1756–1769 (2012)

    Chiang, J., Chen, Y.: Underwater image enhancement by wavelength compensation and dehazing. IEEE Transac- tions on Image Processing 21(4), 1756–1769 (2012)

    work page 2012

  10. [10]

    In: CVPR (2009)

    Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: Imagenet: A large-scale hierarchical image database. In: CVPR (2009)

    work page 2009

  11. [11]

    In: NIPS (2014)

    Eigen, D., Puhrsch, C., Fergus, R.: Depth map prediction from a single image using a multi-scale deep network. In: NIPS (2014)

    work page 2014

  12. [12]

    Enhancing Underwater Imagery using Generative Adversarial Networks

    Fabbri, C., Islam, M.J., Sattar, J.: Enhancing underwa- ter imagery using generative adversarial networks. arXiv preprint arXiv:1801.04011 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  13. [13]

    In: NIPS (2014)

    Goodfellow, I.: Generative adversarial nets. In: NIPS (2014)

    work page 2014

  14. [14]

    In: NIPS (2017)

    Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V., Courville, A.C.: Improved training of wasserstein gans. In: NIPS (2017)

    work page 2017

  15. [15]

    TIP (2018)

    Guo, C., Li, C., Guo, J., etal: Hierarchical features driven residual learning for depth map super-resolution. TIP (2018)

    work page 2018

  16. [16]

    Guo, Y., Li, H., Zhuang, P.: Underwater image enhance- ment using a multiscale dense generative adversarial net- work. IEEE J. Oceanic. Eng. (2019)

    work page 2019

  17. [17]

    TPAMI (2011) 20 Saeed Anwar∗, Chongyi Li ∗

    He, K., Sun, J., Tang, X.: Single image haze removal using dark channel prior. TPAMI (2011) 20 Saeed Anwar∗, Chongyi Li ∗

    work page 2011

  18. [18]

    TPAMI (2013)

    He, K., Sun, J., Tang, X.: Guided image filtering. TPAMI (2013)

    work page 2013

  19. [19]

    In: CVPR (2016)

    He, K., Zhang, X., Ren, S., etal: Deep residual learniing for image recognition. In: CVPR (2016)

    work page 2016

  20. [20]

    In: CVPR, pp

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR, pp. 770–778 (2016)

    work page 2016

  21. [21]

    In: ICIP (2018)

    Hou, M., Liu, R., Fan, X., Luo, Z.: Joint residual learning for underwater image enhancement. In: ICIP (2018)

    work page 2018

  22. [22]

    Ioffe, S., Szegedy, C.: Batch normalization: Accelerating deep network training by reducing internal covariate shift (2015)

    work page 2015

  23. [23]

    In: CVPR (2017)

    Isola, P., Zhu, J.Y., Zhou, T., Efros, A.A.: Image-to- image translation with conditional adversarial networks. In: CVPR (2017)

    work page 2017

  24. [24]

    In: Consumer depth cameras for computer vision (2013)

    Janoch, A., Karayev, S., Jia, Y., Barron, J.T., Fritz, M., Saenko, K., Darrell, T.: A category-level 3d object dataset: Putting the kinect to work. In: Consumer depth cameras for computer vision (2013)

    work page 2013

  25. [26]

    In: ECCV (2016)

    Johnson, J., Alahi, A., Fei-Fei, L.: Perceptual losses for real-time style transfer and super-resolution. In: ECCV (2016)

    work page 2016

  26. [27]

    The relativistic discriminator: a key element missing from standard GAN

    Jolicoeur-Martineau, A.: The relativistic discriminator: a key element missing from standard gan. arXiv preprint arXiv:1807.00734 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  27. [28]

    ICLR (2014)

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. ICLR (2014)

    work page 2014

  28. [29]

    Beitrage zur Physik der freien Atmosphare (1924)

    Koschmieder, H.: Theorie der horizontalen sichtweite. Beitrage zur Physik der freien Atmosphare (1924)

    work page 1924

  29. [30]

    In: ICRA (2014)

    Lai, K., Bo, L., Fox, D.: Unsupervised feature learning for 3d scene labeling. In: ICRA (2014)

    work page 2014

  30. [31]

    LeCun, Y., Bengio, Y., Hinton, G.: Nature (2015)

    work page 2015

  31. [32]

    Pro- ceedings of the IEEE (1998)

    LeCun, Y., Bottou, L., Bengio, Y., Haffner, P.: Gradient- based learning applied to document recognition. Pro- ceedings of the IEEE (1998)

    work page 1998

  32. [33]

    In: CVPR (2017)

    Ledig, C., Wang, Z., Shi, W., Theis, L., Huszar, F., Caballero, J., Cunningham, A., Acosta, A., Aitken, A., Tejani, A., et al.: Photo-realistic single image super- resolution using a generative adversarial network. In: CVPR (2017)

    work page 2017

  33. [34]

    IEEE Signal Processing Letters (2018)

    Li, C., Guo, C., Guo, J.: Emerging from water: Under- water image color correction based on weakly supervised color transfer. IEEE Signal Processing Letters (2018)

    work page 2018

  34. [35]

    arXiv preprint arXiv:1901.05495 (2019)

    Li, C., Guo, C., Ren, W., Cong, R., Hou, J., Kwong, S.: An underwater image enhancement dataset and beyond. arXiv preprint arXiv:1901.05495 (2019)

    work page arXiv 1901

  35. [36]

    In: ECCV (2016)

    Li, C., Wand, M.: Precomputed real-time texture synthe- sis with markovian generative adversarial networks. In: ECCV (2016)

    work page 2016

  36. [37]

    A Fusion Adversarial Underwater Image Enhancement Network with a Public Test Dataset

    Li, H., Li, J., Wang, W.: A Fusion Adversarial Under- water Image Enhancement Network with a Public Test Dataset. arXiv e-prints arXiv:1906.06819 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1906

  37. [38]

    IEEE Robotics and Automation Letters (2018)

    Li, J., Skinner, K.A., Eustice, R.M., Johnson-Roberson, M.: Watergan: Unsupervised generative network to en- able real-time color correction of monocular underwater images. IEEE Robotics and Automation Letters (2018)

    work page 2018

  38. [39]

    Optics & Laser Technology (2019)

    Lu, J., Li, N., Zhang, S., Yu, Z., Zheng, H., Zheng, B.: Multi-scale adversarial network for underwater image restoration. Optics & Laser Technology (2019)

    work page 2019

  39. [40]

    In: NIPS (2016)

    Mao, X., Shen, C., Yang, Y.B.: Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connections. In: NIPS (2016)

    work page 2016

  40. [41]

    Spectral Normalization for Generative Adversarial Networks

    Miyato, T., Kataoka, T., Koyama, M., Yoshida, Y.: Spec- tral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  41. [42]

    In: ICML (2010)

    Nair, V., Hinton, G.E.: Rectified linear units improve re- stricted boltzmann machines. In: ICML (2010)

    work page 2010

  42. [43]

    In: OCEANS (2015)

    Oleari, F., Kallasi, F., Rizzini, D.L., Aleotti, J., Caselli, S.: An underwater stereo vision system: from design to deployment and dataset acquistion. In: OCEANS (2015)

    work page 2015

  43. [44]

    IEEE Jour- nal of Oceanic Engineering (2015)

    Panetta, K., Gao, C., Agaian, S.: Human-visual-system- inspired underwater image quality measures. IEEE Jour- nal of Oceanic Engineering (2015)

    work page 2015

  44. [45]

    Ecology and evolution (2017)

    Pizarro, O., Friedman, A., Bryson, M., Williams, S.B., Madin, J.: A simple, fast, and repeatable survey method for underwater visual 3d benthic mapping and monitor- ing. Ecology and evolution (2017)

    work page 2017

  45. [46]

    Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

    Radford, A., Metz, L., Chintala, S.: Unsupervised rep- resentation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434 (2015)

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  46. [47]

    TPAMI (2017)

    Ren, S., He, K., Girshick, R., etal: Guided image filtering. TPAMI (2017)

    work page 2017

  47. [48]

    In: International Conference on Medical image computing and computer-assisted intervention (2015)

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolu- tional networks for biomedical image segmentation. In: International Conference on Medical image computing and computer-assisted intervention (2015)

    work page 2015

  48. [49]

    In: CVPR (2013)

    Shotton, J., Glocker, B., Zach, C., Izadi, S., Criminisi, A., Fitzgibbon, A.: Scene coordinate regression forests for camera relocalization in rgb-d images. In: CVPR (2013)

    work page 2013

  49. [50]

    In: ECCV (2012)

    Silberman, N., Hoiem, D., Kohli, P., Fergus, R.: Indoor segmentation and support inference from rgbd images. In: ECCV (2012)

    work page 2012

  50. [51]

    ICLR (2014)

    Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. ICLR (2014)

    work page 2014

  51. [52]

    In: CVPRW (2017)

    Skinner, K.A., Johnson-Roberson, M.: Underwater image dehazing with a light field camera. In: CVPRW (2017)

    work page 2017

  52. [53]

    IET Image Processing (2018)

    Sun, X., Liu, L., Li, Q., Dong, J., Lima, E., Yin, R.: Deep pixel to pixel network for underwater image enhancement and restoration. IET Image Processing (2018)

    work page 2018

  53. [54]

    All-In-One Underwater Image Enhancement using Domain-Adversarial Learning

    Uplavikar, P., Wu, Z., Wang, Z.: All-in-one underwater image enhancement using domain-adversarial learning. arXiv preprint arXiv:1905.13342 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  54. [55]

    IEEE Signal Processing Letters (2015)

    Wang, S., Ma, K., Yeganeh, H., Wang, Z., Lin, W.: A patch-structure representation method for quality assess- ment of contrast changed images. IEEE Signal Processing Letters (2015)

    work page 2015

  55. [56]

    In: ICIP (2017)

    Wang, Y., Zhang, J., Cao, Y., Wang, Z.: A deep cnn method for underwater image enhancement. In: ICIP (2017)

    work page 2017

  56. [57]

    IEEE signal processing letters (2002)

    Wang, Z., Bovik, A.C.: A universal image quality index. IEEE signal processing letters (2002)

    work page 2002

  57. [58]

    Synthesis Lectures on Image, Video, and Multimedia Pro- cessing (2006)

    Wang, Z., Bovik, A.C.: Modern image quality assessment. Synthesis Lectures on Image, Video, and Multimedia Pro- cessing (2006)

    work page 2006

  58. [59]

    IEEE signal processing magazine (2009)

    Wang, Z., Bovik, A.C.: Mean squared error: Love it or leave it? a new look at signal fidelity measures. IEEE signal processing magazine (2009)

    work page 2009

  59. [60]

    TIP (2004)

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P., et al.: Image quality assessment: from error visibility to structural similarity. TIP (2004)

    work page 2004

  60. [61]

    In: The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003 (2003)

    Wang, Z., Simoncelli, E.P., Bovik, A.C.: Multiscale struc- tural similarity for image quality assessment. In: The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003 (2003)

    work page 2003

  61. [62]

    In: Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Wu, Q., Wang, P., Shen, C., Dick, A., van den Hengel, A.: Ask me anything: Free-form visual question answering based on knowledge from external sources. In: Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4622–4630 (2016)

    work page 2016

  62. [63]

    TIP (2015) Diving Deeper into Underwater Image Enhancement: A Survey 21

    Yang, M., Sowmya, A.: An underwater color image qual- ity evaluation metric. TIP (2015) Diving Deeper into Underwater Image Enhancement: A Survey 21

    work page 2015

  63. [64]

    In: Pacific Rim Conference on Multimedia (2018)

    Ye, X., Xu, H., Ji, X., Xu, R.: Underwater image enhance- ment using stacked generative adversarial networks. In: Pacific Rim Conference on Multimedia (2018)

    work page 2018

  64. [65]

    In: CVPR (2017)

    Zhang, K., Zuo, W., Gu, S.: Learning deep cnn denoiser prior for image restoration. In: CVPR (2017)

    work page 2017

  65. [66]

    In: ICCV (2017)

    Zhu, Y., Park, T., Efros, A.: Unpaired image-to-image translation using cycle-consistent adversarial networks. In: ICCV (2017)

    work page 2017