Pith. sign in

REVIEW 4 major objections 5 minor 57 references

Physics Informed Capsule Enhanced Variational AutoEncoder for Underwater Image Enhancement

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A dual-stream autoencoder that couples an underwater formation model with capsule clustering restores hazy underwater images to top reported quality at about one-third the compute of diffusion-based competitors.

desk verdict A plausible physics-informed extension of the authors' own CE-VAE with useful ablations, but the abstract's SOTA numbers are not backed by Table 1. read the letter →

arxiv 2506.04753 v1 pith:EVDUACGD submitted 2025-06-05 cs.CV cs.AIeess.IV

Niki Martinel , Rita Pucci This is my paper
classification cs.CVcs.AIeess.IV
keywords underwaterimageenhancementJaffe-McGlamerymodelcapsulenetworksvariationalautoencodertransmissionmapestimationbackgroundlightLaplacianpyramidlossrestoration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that underwater image enhancement improves when the network is explicitly tied to the physics of how light travels through water, rather than learning the correction purely from data. It proposes pi-CE-VAE, a dual-stream autoencoder in which one stream estimates the transmission map and spatially varying background light of the underwater formation model, while a parallel stream uses capsule clustering to capture object-level structure. A parameter-free enhancer then inverts the formation equation to produce the clear image. On six benchmarks, the authors report consistent gains of about +0.5 dB PSNR over the best prior methods while using roughly one-third of their FLOPs, or more than +1 dB when matched in compute. If correct, this would make physics-informed modeling and capsule representations a practical alternative to heavy diffusion or transformer approaches for underwater restoration.

What carries the argument

The load-bearing object is the simplified underwater formation equation $I_{\rm deg} = I_{\rm clear} \odot T + \alpha \odot (1 - T)$, together with the two estimates produced by the physics estimator: a per-pixel transmission map $\hat{T}$ and a spatially varying background-light map $\hat{\alpha}$. The physics enhancer inverts this equation in a parameter-free way, subtracting the backscatter and dividing out the transmission. Around that core, capsule clustering applies routing-by-agreement to the encoder latents to form entity vectors whose presence and pose are folded into the decoder through a residual addition $\hat{X} = X + C$. The four-loss objective enforces that the estimates reproduce the observed degradation, match a derived transmission target, and agree with the clear image at multiple spatial frequencies.

What would settle it

Evaluate pi-CE-VAE on underwater images with measured range maps and attenuation coefficients and check whether the estimated transmission map $\hat{T}$ tracks the physically measured transmission: if the correlation is low, or if removing the cycle and transmission losses leaves PSNR unchanged, the claim that physics guidance drives the gains fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that explicitly enforcing the underwater image formation model inside a capsule-based variational autoencoder yields enhancements that are both more accurate and cheaper than current methods. The network estimates the physical parameters, $\hat{T}$ and $\hat{\alpha}$, from the degraded image and produces the clear image through the rearrangement $\hat{I}_{\rm clear} = (\tilde{I}_{\rm clear} - \hat{\alpha} \odot (1 - \hat{T})) \oslash \hat{T}$, so the inversion itself requires no learned parameters. Entity-level information from capsule clustering is added residually to the latents, preserving spatial detail that pixel-wise CNNs miss. Training combines a reconstruction loss, a Laplacian pyramid loss, a cycle loss that re-synthesizes the degraded image from the estimated physics, and a transmission supervision loss. The authors report the highest PSNR on EUVP, UFO-120, and LSUI and argue that the physics guidance is what transfers across diverse water types.

Load-bearing premise

The load-bearing premise is that the simplified, single-transmission formation equation used in training adequately describes real underwater degradation; the supervision target for that physics is derived from the network's own background-light estimate and the ground-truth clear image, not from independent depth or scattering measurements.

Editorial extensions

If this is right

  • Because the physics enhancer is parameter-free, the same inversion step can be attached to any network that produces transmission and background-light estimates, making the physics component portable beyond this architecture.
  • At equal or better PSNR, pi-CE-VAE runs at 900.57 GFLOPs and 80.71 ms latency, about one-third the cost of PA-Diff, which means the reported gap to diffusion models is closed without iterative sampling.
  • The cycle and transmission losses force the estimated $\hat{T}$ and $\hat{\alpha}$ to reproduce the observed degraded image through the formation equation, so outputs that violate the simplified physics are penalized during training.
  • Because capsule clustering adds entity-level information through a residual path rather than full-resolution attention, the decoder retains fine detail while working at reduced resolution.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial extension: the transmission-supervision target in Eq. (18) is constructed from the network's own background-light estimate and the ground-truth clear image, so the physics constraint enforces consistency with the simplified model, not with independent measurements of depth, attenuation, or scattering.
  • Editorial extension: the one-third-FLOP claim depends on how capsule routing iterations and the physics estimator's overhead are counted; an instrumented complexity profile of the full forward pass would make the efficiency comparison directly auditable.
  • Editorial extension: replacing the single transmission term with wavelength-dependent attenuation, or adding a depth head, would let the same architecture output range estimates, connecting underwater enhancement to 3D reconstruction and robot navigation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes pi-CE-VAE, a dual-stream variational autoencoder for underwater image enhancement that combines a capsule-clustering feature stream with a physics estimator producing transmission maps and background light. The enhancement output is obtained by inverting a simplified Jaffe-McGlamery formation model. Training uses reconstruction, Laplacian-pyramid, cycle-consistency, and transmission-supervision losses. Experiments are reported on six datasets (EUVP, UFO120, LSUI, U45, SQUID, UCCS), with ablations over capsule integration, physics-enhancer use, and loss components. The central claims are consistent +0.5 dB PSNR improvements over the best existing methods, one-third computational cost of the best methods, and more than +1 dB PSNR improvement when compared with methods of similar computational budget.

Significance. If the quantitative claims were correct, the paper would provide a meaningful efficiency-quality trade-off for underwater image enhancement, with the capsule-physics hybrid architecture being a novel combination. The experimental design includes ablations that isolate the capsule fusion mechanism, the physics enhancer, and the loss components, which is a strength. However, the headline performance margins are not supported by the paper's own tables, and the physics losses are partly self-referential, so the claimed advantage over the state of the art is not established. The contribution is potentially salvageable as an incremental improvement, but the current presentation significantly overstates the results.

major comments (4)
  1. [Abstract & Section 5.4.1, Table 1] The abstract and Section 5.4.1 claim 'consistent improvements of +0.5dB PSNR over the best existing methods.' Table 1 does not support this: the improvements over the best prior method are +0.44 dB on EUVP (28.91 vs. PA-Diff 28.47), +0.05 dB on UFO-120 (26.53 vs. PA-Diff 26.48), and +0.04 dB on LSUI (27.81 vs. DM-Underwater 27.77). None of the three datasets reaches +0.5 dB, and the average margin is about +0.18 dB. This arithmetic inconsistency directly undermines the primary empirical contribution of the paper.
  2. [Abstract & Section 5.4.1, Table 1] The abstract states that the method requires 'only one-third of their computational complexity (FLOPs).' Table 1 reports pi-CE-VAE at 900.57 GFLOPs and PA-Diff at 3.65 TFLOPs, a ratio of 4.05x, i.e., approximately one-quarter of the FLOPs, not one-third. Section 5.4.1 itself says '4.3× reduction,' which is also inconsistent with the abstract's 'one-third.' The computational-efficiency claim should be restated accurately.
  3. [Eqs. (17), (18), Section 4.6] The 'physics-informed' supervision is partly self-referential. The transmission-map target in Eq. (18) is computed as T = (Ideg - alpha_hat) / (Iclear - alpha_hat + epsilon), where alpha_hat is the network's own background-light estimate, so the target is not an independent measurement of depth or scattering. Similarly, the cycle loss in Eq. (17) reuses the same simplified Jaffe-McGlamery model that the enhancer in Eq. (11) inverts. Consequently, the physics losses enforce self-consistency with the assumed model rather than providing external physical grounding. The claim that the method 'respects underwater formation constraints' and the expected generalization to real-world scenes would be strengthened by an experiment using independent transmission/backscatter measurements or synthetic data with known ground-truth physical parameters.
  4. [Abstract & Section 5.4.1] The alternative claim of 'more than +1dB PSNR improvement when compared to methods with similar computational budgets' is vague: no definition of 'similar computational budget' is given, and the comparison set is not identified. Against DM-Underwater (1.34 TFLOPs), the LSUI improvement is only +0.04 dB; against the authors' own CE-VAE (473.77 GFLOPs), the compute is not similar (roughly 1.9x higher). The claim should be made precise, with a stated reference class, or removed.
minor comments (5)
  1. [Section 5.2] The metrics paragraph lists LPIPS as a full-reference metric, but Tables 1 and 2 report CLIP-IQA instead; LPIPS results never appear. Please reconcile the metrics list with the reported tables.
  2. [Eq. (14)] Equation (14) defines lambda_k(I) but the expressions use G_k(Iclear) instead of G_k(I); this is a typo that makes the definition inconsistent with Eq. (13).
  3. [Eq. (1), Section 3] The notation '0α' appears for the background light in the text above Eq. (1); it should be 'α' or a properly defined symbol.
  4. [Throughout] There are several typographical and formatting issues: 'asscociated' in Eq. (13), 'T able' in several table captions, 'FLOPS' vs. 'FLOPs' inconsistency, and the 'check-circle' symbols in Table 4. These should be cleaned up.
  5. [Section 5.6, Figure 7] The qualitative comparison is based on '5 random images,' but no selection protocol is described. For a reproducibility-focused paper, a deterministic or stratified selection procedure would be preferable.

Circularity Check

1 steps flagged · score 4.0 of 10

Transmission-map supervision in Eq. (18) is a self-consistency target built from the network's own background-light estimate; the central benchmark claims remain externally grounded.

  1. self definitional [Section 4.6.4, Eqs. (18)-(19)]
    "Since we have two unknowns in (1), we can rearrange (1) while considering our estimate for the backscatter light bα to obtain the expected transmission map T = (Ideg − bα) ⊘ (Iclear − bα + ε), (18) where ε is a small constant to avoid division by zero. The transmission supervision loss is then defined by Ltransmission = ∥T − bT∥1. (19)"

    The target T in Eq. (18) is not an independent physical measurement: it is the algebraic rearrangement of the assumed Jaffe-McGlamery model (1) evaluated with bα = Φ(Ideg), the network's own background-light prediction. Supervising bT with this T therefore only forces bT and bα to be mutually consistent under the same equation that the enhancer (Eq. 11) inverts; it cannot validate the physical correctness of either estimate. Thus the 'physics-guided' transmission prediction reduces, by construction, to self-consistency among the network's own outputs under an assumed forward model.

full rationale

The central performance claim is measured against external ground-truth images on EUVP, UFO-120, LSUI and non-reference benchmarks with standard metrics, so it does not reduce to the network's own outputs. However, the physics-informed part is partially self-referential: the transmission-map target in Eq. (18) is computed from the network's own background-light estimate bα and the same simplified Jaffe-McGlamery equation used by the enhancer (Eq. 11) and the cycle loss (Eqs. 16-17). This makes the 'physical adherence' a consistency constraint rather than independent physical supervision. Separately, the abstract's '+0.5dB over best existing methods' is not actually supported by Table 1 (margins are +0.44dB on EUVP, +0.05dB on UFO-120, and +0.04dB on LSUI), and the '>+1dB at similar compute' claim is only realized against the authors' own CE-VAE, which has roughly half the FLOPs (473.77G vs 900.57G); these are empirical-claim problems, not circular derivations. Because the headline SOTA result still depends on external benchmarks, the circularity is partial and non-central, warranting a modest score rather than a 6+.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper's physics guidance rests on an assumed formation model and on loss targets that are not independent of the network's own estimates. The final benchmark comparisons provide external grounding, but the physical quantities (transmission, background light) are never validated against measured depth or scattering data.

free parameters (3)
  • physics loss weight eta = 0.0001
    Eq (20); chosen by hand to balance cycle and transmission losses. No sensitivity analysis is reported.
  • Laplacian pyramid level weights omega_k = omega_k = 1/2^k with L = 3
    Section 4.6.2 and Table 4; exponential weighting and 3 levels selected by ablation.
  • epsilon in transmission target = unspecified small constant
    Eq (18); added to avoid division by zero, affecting the transmission supervision target when Iclear - alpha_hat is near zero.
assumptions (4)
  • domain assumption The simplified Jaffe-McGlamery model, Ideg = Iclear * T + alpha * (1 - T), adequately describes degradation in the training and test datasets.
    Used in Eq (1), Eq (10), Eq (16), and the enhancer Eq (11). No external validation of T or alpha is provided.
  • domain assumption Forward scattering is negligible at typical underwater imaging distances.
    Section 3 states forward scattering is omitted. If it is not negligible, the inverse enhancement may be systematically wrong.
  • domain assumption A single per-pixel transmission coefficient T applies equally to all color channels.
    Eqs (1) and (2) use scalar T. Real water attenuation is wavelength-dependent, and the paper does not model that in the physics estimator.
  • ad hoc to paper The transmission target computed in Eq (18) from the model's own alpha_hat and the ground-truth clear image provides useful supervision.
    L_transmission enforces agreement with an algebraically derived T that is not an independent measurement of transmission or depth.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics Informed Capsule Enhanced Variational AutoEncoder for Underwater Image Enhancement." pith.science (2026). https://pith.science/paper/EVDUACGD

@misc{pith2026250604753,
  author       = {Pith},
  title        = {Pith review of: Physics Informed Capsule Enhanced Variational AutoEncoder for Underwater Image Enhancement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EVDUACGD}},
  note         = {Machine review of arXiv:2506.04753}
}
abstract

We present a novel dual-stream architecture that achieves state-of-the-art underwater image enhancement by explicitly integrating the Jaffe-McGlamery physical model with capsule clustering-based feature representation learning. Our method simultaneously estimates transmission maps and spatially-varying background light through a dedicated physics estimator while extracting entity-level features via capsule clustering in a parallel stream. This physics-guided approach enables parameter-free enhancement that respects underwater formation constraints while preserving semantic structures and fine-grained details. Our approach also features a novel optimization objective ensuring both physical adherence and perceptual quality across multiple spatial frequencies. To validate our approach, we conducted extensive experiments across six challenging benchmarks. Results demonstrate consistent improvements of $+0.5$dB PSNR over the best existing methods while requiring only one-third of their computational complexity (FLOPs), or alternatively, more than $+1$dB PSNR improvement when compared to methods with similar computational budgets. Code and data \textit{will} be available at https://github.com/iN1k1/.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 39 canonical work pages

  1. [1]

    A., Mindell, D., Roman, C., Sakellariou, D.: Robotic tools for deep water archae- ology: Surveying an ancient shipwreck with an autonomous underwater vehicle

    Bingham, B., Foley, B., Singh, H., Camilli, R., Delaporta, K., Eustice, R., Mallios, 16 Figure 7: Enhanced images comparison on five random samples taken from the vali- dation set of the three considered full-reference datasets. A., Mindell, D., Roman, C., Sakellariou, D.: Robotic tools for deep water archae- ology: Surveying an ancient shipwreck with an ...

    doi:10.1002/rob 2010

  2. [2]

    In: IROS, pp

    Shkurti, F., Xu, A., Meghjani, M., Higuera, J.C.G., Girdhar, Y., Giguere, P., Dey, B.B., Li, J., Kalmbach, A., Prahacs, C., et al.: Multi-domain monitoring of marine environments using a heterogeneous robot team. In: IROS, pp. 1747–1753 (2012). https://doi.org/10.1109/IROS.2012.6385685

    work page arXiv 2012

  3. [3]

    In: Robotics Research, pp

    Whitcomb, L., Yoerger, D.R., Singh, H., Howland, J.: Advances in underwa- ter robot vehicles for deep ocean exploration: Navigation, control, and survey operations. In: Robotics Research, pp. 439–448 (2000). https://doi.org/10.1007/ 978-1-4471-0765-1 53

    work page 2000

  4. [4]

    IEEE Journal of Oceanic Engineering15(2), 101–111 (1990) https:// doi.org/10.1109/48.50695

    Jaffe, J.S.: Computer modeling and the design of optimal underwater imaging systems. IEEE Journal of Oceanic Engineering15(2), 101–111 (1990) https:// doi.org/10.1109/48.50695

    doi:10.1109/48.50695 1990

  5. [5]

    SIO ref 75(2) (1975) 17

    McGlamery, B.: Computer analysis and simulation of underwater camera system performance. SIO ref 75(2) (1975) 17

    work page 1975

  6. [6]

    IEEE Transactions on Image Processing 25(12), 5664–5677 (2016) https://doi.org/10.1109/TIP.2016.2612882

    Li, C.-Y., Guo, J.-C., Cong, R.-M., Pang, Y.-W., Wang, B.: Underwater image enhancement by dehazing with minimum information loss and histogram distri- bution prior. IEEE Transactions on Image Processing 25(12), 5664–5677 (2016) https://doi.org/10.1109/TIP.2016.2612882

    work page arXiv 2016

  7. [7]

    Applied soft computing27, 219–230 (2015) https://doi.org/10.1016/j.asoc.2014.11.020

    Ghani, A.S.A., Isa, N.A.M.: Underwater image quality enhancement through inte- grated color model with rayleigh distribution. Applied soft computing27, 219–230 (2015) https://doi.org/10.1016/j.asoc.2014.11.020

    work page doi:10.1016/j.asoc.2014.11.020 2015

  8. [8]

    Applied Optics56(23), 6631–6638 (2017) https://doi.org/10

    Han, P., Liu, F., Yang, K., Ma, J., Li, J., Shao, X.: Active underwater descattering and image recovery. Applied Optics56(23), 6631–6638 (2017) https://doi.org/10. 1364/AO.56.006631

    work page 2017

Show all 57 references
  1. [9]

    Instrumentation Viewpoint (20), 53–54 (2018)

    Neumann, L., Garcia, R., J´ anosik, J., Gracias, N.: Fast underwater color correction using integral images. Instrumentation Viewpoint (20), 53–54 (2018)

    2018

  2. [10]

    Journal of Marine Science and Engineering9(7), 691 (2021) https://doi.org/10.3390/jmse9070691

    Hu, K., Zhang, Y., Weng, C., Wang, P., Deng, Z., Liu, Y.: An underwater image enhancement algorithm based on generative adversarial network and natural im- age quality evaluation index. Journal of Marine Science and Engineering9(7), 691 (2021) https://doi.org/10.3390/jmse9070691

    2021 doi

  3. [11]

    IEEE Robotics and Automation Letters5(2), 3227–3234 (2020) https://doi.org/10.1109/LRA.2020.2974710

    Islam, M.J., Xia, Y., Sattar, J.: Fast underwater image enhancement for improved visual perception. IEEE Robotics and Automation Letters5(2), 3227–3234 (2020) https://doi.org/10.1109/LRA.2020.2974710

    2020

  4. [12]

    Journal of Marine Science and Engineering 7(7), 200 (2019) https://doi.org/10.3390/jmse7070200

    Park, J., Han, D.K., Ko, H.: Adaptive weighted multi-discriminator cyclegan for underwater image enhancement. Journal of Marine Science and Engineering 7(7), 200 (2019) https://doi.org/10.3390/jmse7070200

    2019 doi

  5. [13]

    In: ICRA, pp

    Fabbri, C., Islam, M.J., Sattar, J.: Enhancing underwater imagery using gener- ative adversarial networks. In: ICRA, pp. 7159–7165 (2018). https://doi.org/10. 1109/ICRA.2018.8460552

    2018

  6. [14]

    IET Image Processing15(9), 2010–2019 (2021) https://doi

    Zhang, H., Sun, L., Wu, L., Gu, K.: Dugan: An effective framework for underwater image enhancement. IET Image Processing15(9), 2010–2019 (2021) https://doi. org/10.1049/ipr2.12172

    2021 doi

  7. [15]

    In: International Conference on Computer Vision, pp

    Zhu, J.-Y., Park, T., Isola, P., Efros, A.A.: Unpaired image-to-image transla- tion using cycle-consistent adversarial networks. In: International Conference on Computer Vision, pp. 2223–2232 (2017)

    2017

  8. [16]

    arXiv:2002.01155 (2020) https://doi.org/10.48550/arXiv.2002.01155

    Islam, M.J., Luo, P., Sattar, J.: Simultaneous enhancement and super-resolution of underwater imagery for improved visual perception. arXiv:2002.01155 (2020) https://doi.org/10.48550/arXiv.2002.01155

    2020 doi arXiv

  9. [17]

    IEEE Journal of Oceanic Engineering 18 45(3), 862–870 (2019) https://doi.org/10.1109/JOE.2019.2911447

    Guo, Y., Li, H., Zhuang, P.: Underwater image enhancement using a multi- scale dense generative adversarial network. IEEE Journal of Oceanic Engineering 18 45(3), 862–870 (2019) https://doi.org/10.1109/JOE.2019.2911447

    2019

  10. [18]

    Journal of Marine Science and Engineering 13(5), 899 (2025)

    Ma, Y., Cheng, Y., Zhang, D.: Comparative analysis of traditional and deep learn- ing approaches for underwater remote sensing image enhancement: A quantitative study. Journal of Marine Science and Engineering 13(5), 899 (2025)

    2025

  11. [19]

    In: CMM’06, p

    Bazeille, S., Quidu, I., Jaulin, L., Malkasse, J.-P.: Automatic underwater image pre-processing. In: CMM’06, p. (2006)

    2006

  12. [20]

    IEEE Transactions on Image Processing 27(1), 379–393 (2018) https://doi.org/10.1109/TIP.2017.2759252

    Ancuti, C.O., Ancuti, C., De Vleeschouwer, C., Bekaert, P.: Color balance and fu- sion for underwater image enhancement. IEEE Transactions on Image Processing 27(1), 379–393 (2018) https://doi.org/10.1109/TIP.2017.2759252

    2018

  13. [21]

    In: IEEE International Conference on Image Processing, pp

    Lu, H., Li, Y., Serikawa, S.: Underwater image enhancement using guided trigono- metric bilateral filter and fast automatic color correction. In: IEEE International Conference on Image Processing, pp. 3412–3416 (2013). IEEE

    2013

  14. [22]

    In: IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), pp

    Li, C., Quo, J., Pang, Y., Chen, S., Wang, J.: Single underwater image restoration by blue-green channels dehazing and red channel correction. In: IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 1731–1735 (2016). IEEE

    2016

  15. [23]

    IEICE Transactions on Information and Systems 100(7), 1475–1483 (2017)

    Park, D., Han, D.K., Ko, H.: Enhancing underwater color images via optical imag- ing model and non-local means denoising. IEICE Transactions on Information and Systems 100(7), 1475–1483 (2017)

    2017

  16. [24]

    IEEE Journal of Oceanic Engineering 47(3), 718–735 (2022) https://doi.org/10.1109/JOE.2022

    Zhang, W., Wang, Y., Li, C.: Underwater image enhancement by attenuated color channel correction and detail preserved contrast enhancement. IEEE Journal of Oceanic Engineering 47(3), 718–735 (2022) https://doi.org/10.1109/JOE.2022. 3140563

    2022 doi

  17. [25]

    IEEE Transactions on Image Processing 26(4), 1579–1594 (2017)

    Peng, Y.-T., Cosman, P.C.: Underwater image restoration based on image blur- riness and light absorption. IEEE Transactions on Image Processing 26(4), 1579–1594 (2017)

    2017

  18. [26]

    Optics Communications 574, 130942 (2025)

    Lin, S., Ning, Z., Zhang, R.: Modified optical model and optimized contrast for underwater image restoration. Optics Communications 574, 130942 (2025)

    2025

  19. [27]

    Water 16(5), 626 (2024)

    Wang, R., Zhang, Y., Zhang, Y.: A lightweight multi-branch context network for unsupervised underwater image restoration. Water 16(5), 626 (2024)

    2024

  20. [28]

    Scientific Reports 15(1), 6076 (2025)

    Yan, J., Hu, H., Wang, Y., Nawaz, M.W., Ur Rehman Junejo, N., Guo, E., Feng, H.: Underwater image enhancement via multiscale disentanglement strategy. Scientific Reports 15(1), 6076 (2025)

    2025

  21. [29]

    International Journal of Computer Vision (2023) https://doi.org/10

    Zhou, J., Liu, Q., Jiang, Q., Ren, W., Lam, K.M., Zhang, W.: Underwater camera: Improving visual perception via adaptive dark pixel prior and color 19 correction. International Journal of Computer Vision (2023) https://doi.org/10. 1007/s11263-023-01853-3

    2023

  22. [30]

    In: International Conference on Computer Vision and Pattern Recognitino (CVPR), pp

    Akkaynak, D., Treibitz, T.: Sea-thru: A method for removing water from un- derwater images. In: International Conference on Computer Vision and Pattern Recognitino (CVPR), pp. 1682–1691 (2019)

    2019

  23. [31]

    In: Medical Image Computing and Computer-assisted intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, pp

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-assisted intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, p...

    2015

  24. [32]

    IEEE Transactions on Image Processing 29 (2020)

    Li, C., Guo, C., Ren, W., Cong, R., Hou, J., Kwong, S., Tao, D.: An underwater image enhancement benchmark dataset and beyond. IEEE Transactions on Image Processing 29 (2020)

    2020

  25. [33]

    In: 2022 IEEE International Conference on Unmanned Systems (ICUS) (2022)

    Xing, Z., Cai, M., Li, J.: Improved shallow-uwnet for underwater image enhance- ment. In: 2022 IEEE International Conference on Unmanned Systems (ICUS) (2022)

    2022

  26. [34]

    IEEE Transactions on Broadcasting (2022)

    Qiao, N., Dong, L., Sun, C.: Adaptive deep learning network with multi- scale and multi-dimensional features for underwater image enhancement. IEEE Transactions on Broadcasting (2022)

    2022

  27. [35]

    IEEE Transactions on Image Processing 32, 3066–3079 (2023) https: //doi.org/10.1109/TIP.2023.3276332

    Peng, L., Zhu, C., Bian, L.: U-shape transformer for underwater image en- hancement. IEEE Transactions on Image Processing 32, 3066–3079 (2023) https: //doi.org/10.1109/TIP.2023.3276332

    2023

  28. [36]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp

    Khan, R., Mishra, P., Mehta, N., Phutke, S.S., Vipparthi, S.K., Nandi, S., Murala, S.: Spectroformer: Multi-domain query cascaded transformer network for under- water image enhancement. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 1...

    2024

  29. [37]

    Electronics 12(5), 1227 (2023)

    Han, J., Zhou, J., Wang, L., Wang, Y., Ding, Z.: Fe-gan: Fast and efficient un- derwater image enhancement model based on conditional gan. Electronics 12(5), 1227 (2023)

    2023

  30. [38]

    In: ACM International Conference on Multimedia, pp

    Tang, Y., Kawasaki, H., Iwaguchi, T.: Underwater image enhancement by transformer-based diffusion model with non-uniform sampling for skip strat- egy. In: ACM International Conference on Multimedia, pp. 5419–5427 (2023). https://doi.org/10.1145/3581783.3612378

    2023

  31. [39]

    In: IEEE/CVF International Conference on Computer Vision and Pattern Recognition (2024)

    Zhao, C., Cai, W., Dong, C., Hu, C.: Wavelet-based fourier information inter- action with frequency diffusion adjustment for underwater image restoration. In: IEEE/CVF International Conference on Computer Vision and Pattern Recognition (2024). https://github.com/zhihefang/WF-Diff. 20

    2024

  32. [40]

    Zhao, C., Dong, C., Cai, W.: Learning a physical-aware diffusion model based on transformer for underwater image enhancement (2024)

    2024

  33. [41]

    In: IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV) (2025)

    Pucci, R., Martinel, N.: Ce-vae: Capsule enhanced variational autoencoder for un- derwater image enhancement. In: IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV) (2025). https://github.com/iN1k1/

    2025

  34. [42]

    In: Advances in Neural Information Processing Systems, pp

    Sabour, S., Frosst, N., Hinton, G.E.: Dynamic routing between capsules. In: Advances in Neural Information Processing Systems, pp. 3856–3866 (2017)

    2017

  35. [43]

    In: International Conference on Computer Vision and Pattern Recognition, pp

    Esser, P., Rombach, R., Ommer, B.: Taming transformers for high-resolution image synthesis. In: International Conference on Computer Vision and Pattern Recognition, pp. 12873–12883 (2021)

    2021

  36. [44]

    In: International Conference on Multimedia Modeling, pp

    Huang, D., Wang, Y., Song, W., Sequeira, J., Mavromatis, S.: Shallow-water image enhancement using relative global histogram stretching based on adaptive parameter acquisition. In: International Conference on Multimedia Modeling, pp. 453–465 (2018). https://doi.org/10.1007/978...

    2018 doi

  37. [45]

    IEEE Computer Graphics and Applications 36, 24–35 (2016) https://doi.org/10.1109/MCG.2016

    Drews, P.L.J., Nascimento, E.R., Botelho, S.S.C., Campos, M.F.M.: Underwater depth estimation and image restoration based on single images. IEEE Computer Graphics and Applications 36, 24–35 (2016) https://doi.org/10.1109/MCG.2016. 26

    2016 doi

  38. [46]

    IEEE Transactions on Image Processing 26, 1579–1594 (2017) https://doi.org/10.1109/TIP.2017.2663846

    Peng, Y.T., Cosman, P.C.: Underwater image restoration based on image blurri- ness and light absorption. IEEE Transactions on Image Processing 26, 1579–1594 (2017) https://doi.org/10.1109/TIP.2017.2663846

    2017

  39. [47]

    IEEE Transactions on Circuits and Systems for Video Technology 33, 2561–2576 (2023) https://doi.org/10.1109/TCSVT.2022.3225376

    Li, K., Wu, L., Qi, Q., Liu, W., Gao, X., Zhou, L., Song, D.: Beyond single reference for training: Underwater image enhancement via comparative learning. IEEE Transactions on Circuits and Systems for Video Technology 33, 2561–2576 (2023) https://doi.org/10.1109/TCSVT.2022.3225376

    2023

  40. [48]

    Zhiying Jiang, Fan, X.: Twin adversarial contrastive learning for underwater image enhancement and beyond

    Risheng Liu, H.Y. Zhiying Jiang, Fan, X.: Twin adversarial contrastive learning for underwater image enhancement and beyond. In: IEEE Transactions on Image Processing (2022). IEEE

    2022

  41. [49]

    Sensors 18(9), 3153 (2018) https://doi.org/10.3390/s18093153

    Deng, F., Pu, S., Chen, X., Shi, Y., Yuan, T., Pu, S.: Hyperspectral image classi- fication with capsule network using limited training samples. Sensors 18(9), 3153 (2018) https://doi.org/10.3390/s18093153

    2018 doi

  42. [50]

    IEEE Transactions on Circuits and Systems for Video Technology 30(12), 4861–4875 (2020)

    Liu, R., Fan, X., Zhu, M., Hou, M., Luo, Z.: Real-world underwater enhancement: Challenges, benchmarks, and solutions under natural light. IEEE Transactions on Circuits and Systems for Video Technology 30(12), 4861–4875 (2020)

    2020

  43. [51]

    arXiv preprint arXiv:1906.06819 (2019) 21

    Li, H., Li, J., Wang, W.: A fusion adversarial underwater image enhancement network with a public test dataset. arXiv preprint arXiv:1906.06819 (2019) 21

    2019 arXiv

  44. [52]

    IEEE transactions on pattern analysis and machine intelligence 43(8), 2822–2837 (2020)

    Berman, D., Levy, D., Avidan, S., Treibitz, T.: Underwater single image color restoration using haze-lines and a new quantitative dataset. IEEE transactions on pattern analysis and machine intelligence 43(8), 2822–2837 (2020)

    2020

  45. [53]

    IEEE Transactions on Image Processing 13(4), 600–612 (2004) https://doi.org/10.1109/TIP.2003.819861

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P.: Image quality assess- ment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13(4), 600–612 (2004) https://doi.org/10.1109/TIP.2003.819861

    2004

  46. [54]

    In: International Conference on Computer Vision and Pattern Recognition, pp

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: International Conference on Computer Vision and Pattern Recognition, pp. 586–595 (2018). https://doi. org/10.1109/cvpr.2018.00068

    2018

  47. [55]

    IEEE Transactions on Image Processing24(12), 6062–6071 (2015) https://doi

    Yang, M., Sowmya, A.: An underwater color image quality evaluation metric. IEEE Transactions on Image Processing24(12), 6062–6071 (2015) https://doi. org/10.1109/TIP.2015.2491020

    2015

  48. [56]

    IEEE Journal of Oceanic Engineering 41(3), 541–551 (2016) https://doi.org/10.1109/JOE.2015.2469915

    Panetta, K., Gao, C., Agaian, S.: Human-visual-system-inspired underwater im- age quality measures. IEEE Journal of Oceanic Engineering 41(3), 541–551 (2016) https://doi.org/10.1109/JOE.2015.2469915

    2016

  49. [57]

    In: AAAI (3)

    Wang, J., Chan, K.C.K., Loy, C.C.: Exploring clip for assessing the look and feel of images. In: AAAI (3). http://arxiv.org/abs/2207.12396 22

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.