pith. machine review for the scientific record. sign in

arxiv: 2605.10374 · v1 · submitted 2026-05-11 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Halo Separation-guided Underwater Multi-scale Image Restoration

Jiaxin Yang , Honglin Liu , Yongli Wang , Shuyi Cao , Chengcheng Jiang , Jiale Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:29 UTC · model grok-4.3

classification 💻 cs.CV
keywords underwater image restorationhalo separationgradient minimizationmulti-scale recoveryartificial light correctionAUV imagingimage enhancement
0
0 comments X

The pith

An iterative network separates halos via gradient minimization then recovers masked details at multiple scales to restore underwater images.

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

The paper sets out to show that halos from artificial lights in underwater AUV captures can be corrected by first isolating the halo layer through gradient minimization and then restoring the hidden scene content with multi-scale processing. This matters because existing enhancement techniques produce poor results when lights create foreground halos, degrading follow-on tasks such as detection or navigation. The method trains on synthetic halo examples drawn from standard underwater datasets and adds a radial gradient constraint derived from observed brightness patterns. If the approach succeeds, restored images retain more structural detail and color fidelity in real artificial-light scenes.

Core claim

The central claim is that an iterative two-part network, consisting of a halo layer separation sub-network driven by gradient minimization and a multi-scale recovery sub-network, can isolate and remove artificial-light halos while reconstructing the underlying image information, yielding higher-quality restorations than prior underwater enhancement methods when tested on real underwater halo images.

What carries the argument

The iterative network formed by a halo layer separation sub-network that applies gradient minimization to isolate the halo and a multi-scale recovery sub-network that reconstructs information masked by the halo.

If this is right

  • Underwater image enhancement becomes more robust specifically under artificial illumination conditions.
  • Downstream AUV vision tasks receive higher-quality input images with fewer light-induced degradations.
  • Brightness distribution analysis supplies a radial gradient constraint that further aids halo removal.
  • The separation-then-recovery structure provides a modular template for handling other localized degradations in underwater scenes.

Where Pith is reading between the lines

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

  • The same separation logic could be adapted to correct similar bright-spot artifacts in other scattering media such as fog or turbid water.
  • Combining the halo separation stage with existing color-correction modules might compound gains in low-visibility environments.
  • Real-time variants of the iterative structure could support live navigation and inspection tasks on AUVs.

Load-bearing premise

Training the network exclusively on synthetic halo images generated from UIEB and EUVP datasets will enable it to remove halos from genuine underwater images captured with real artificial lights.

What would settle it

The method produces restored images that still show visible halo artifacts or lower quality scores when evaluated on a fresh set of real underwater photographs containing artificial light sources not present in the training data.

read the original abstract

Underwater images captured by Autonomous Underwater Vehicles (AUVs) are inevitably affected by artificial light sources, which often produce halos in the foreground of the camera and seriously interfere with the quality of the image. The existing underwater image enhancement methods fail to fully consider this key problem, and the robustness of processing images under artificial light scenes is poor. In practical applications, since underwater image enhancement itself is a very challenging task, the influence of artificial light sources will lead to serious degradation of image performance and affect subsequent vision tasks. In order to effectively deal with this problem, this paper designs a single halo image correction method based on an iterative structure. The network is mainly divided into two sub-networks, one is the halo layer separation sub-network which aims to separate the halo by gradient minimization, and the other is the multi-scale recovery sub-network which aims to recover the image information masked by halo. The UIEB and EUVP synthetic datasets are used for training to ensure that the network can fully learn the characteristics and laws of underwater halo images. Then a large number of halo images taken in an underwater environment with real artificial light are collected for testing. In addition, the brightness distribution characteristics of underwater halo images are analyzed and the radial gradient is introduced to constraint eliminate halo to improve the effect of underwater image restoration.

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 paper proposes an iterative network for underwater image restoration under artificial light halos, consisting of a halo layer separation sub-network (using gradient minimization and radial gradient constraints derived from brightness distribution analysis) and a multi-scale recovery sub-network. It is trained on synthetic halo images generated from the UIEB and EUVP datasets and tested on real halo images collected in underwater environments with artificial lights, claiming improved correction of halos and recovery of masked scene content compared to existing enhancement methods.

Significance. If the synthetic-to-real generalization holds with supporting evidence, the work addresses a practical gap in underwater vision for AUVs by targeting halo artifacts from artificial sources, which current methods overlook. The incorporation of domain-specific constraints like radial gradients and the iterative design could offer a targeted improvement, provided quantitative validation confirms it outperforms baselines without introducing distortions.

major comments (2)
  1. [Abstract] Abstract: The central claim is that the iterative halo-separation network trained on synthetic halos from UIEB/EUVP 'effectively corrects halo images and improves underwater image restoration quality' on real artificial-light halo images collected for testing. However, the manuscript provides no quantitative results (e.g., no-reference metrics such as UIQM or UCIQE), no visual comparisons, no ablation studies, and no baseline comparisons on those real test images, which is load-bearing for validating the synthetic-to-real transfer and the effectiveness for AUV applications.
  2. [Method] Method description (halo separation sub-network): The approach relies on gradient minimization plus a radial gradient constraint to isolate halos, assuming synthetic additive halos model real scattering and intensity profiles. No evidence is given that this holds for real data (e.g., differing turbidity interactions or non-additive effects), and without reported metrics or failure-case analysis on the collected real images, the assumption remains untested and risks residual halos or foreground distortion.
minor comments (2)
  1. [Datasets and Training] The description of how synthetic halos are generated from UIEB and EUVP (e.g., exact parameters for halo addition) is not detailed enough for reproducibility; a supplementary section or equation would help.
  2. [Overall] Notation for the iterative structure and sub-networks could be clarified with a diagram or pseudocode to distinguish the gradient-minimization step from the recovery step.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which underscores the importance of rigorous validation for synthetic-to-real transfer in underwater image restoration. We have revised the manuscript to strengthen the evidence on real test images while preserving the core contributions. Our point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim is that the iterative halo-separation network trained on synthetic halos from UIEB/EUVP 'effectively corrects halo images and improves underwater image restoration quality' on real artificial-light halo images collected for testing. However, the manuscript provides no quantitative results (e.g., no-reference metrics such as UIQM or UCIQE), no visual comparisons, no ablation studies, and no baseline comparisons on those real test images, which is load-bearing for validating the synthetic-to-real transfer and the effectiveness for AUV applications.

    Authors: We agree that quantitative and visual evaluations on the real test images are essential to substantiate the claims of effective halo correction and improved restoration. In the revised manuscript, we now report no-reference metrics (UIQM and UCIQE) computed across the collected real halo images, provide additional side-by-side visual comparisons with existing enhancement baselines, and include ablation studies that isolate the contribution of the halo separation and multi-scale recovery components on real data. These additions directly address the need for evidence of synthetic-to-real generalization and practical utility for AUV applications. revision: yes

  2. Referee: [Method] Method description (halo separation sub-network): The approach relies on gradient minimization plus a radial gradient constraint to isolate halos, assuming synthetic additive halos model real scattering and intensity profiles. No evidence is given that this holds for real data (e.g., differing turbidity interactions or non-additive effects), and without reported metrics or failure-case analysis on the collected real images, the assumption remains untested and risks residual halos or foreground distortion.

    Authors: The referee rightly notes that the additive halo modeling assumption and the effectiveness of the gradient minimization plus radial gradient constraint require explicit validation on real data. Although the radial gradient constraint was derived from brightness distribution analysis of observed underwater halos, we acknowledge that real scattering can involve non-additive interactions not fully captured in synthesis. In the revision, we have added a limitations discussion with failure-case examples on real images (showing minor residual halos under high turbidity), reported the no-reference metrics on real data to quantify overall performance, and included visual results of the separated halo layers on real test images to demonstrate the separation behavior. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture and constraints derived from external analysis and standard training

full rationale

The paper derives its iterative halo-separation network (gradient minimization plus radial gradient constraint) and multi-scale recovery sub-network from an analysis of underwater halo brightness distributions, then trains the resulting model on synthetic halos added to UIEB/EUVP images before testing on separately collected real images. No equation or claim reduces by construction to a fitted parameter renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled via self-citation. The central design choices remain independent of the target real-halo test set.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim depends on these domain assumptions about separability and data fidelity, as no independent verification is described.

axioms (2)
  • domain assumption Gradient minimization can effectively separate the halo layer in underwater images
    This is the aim of the halo layer separation sub-network.
  • domain assumption Synthetic datasets sufficiently represent real-world underwater halo characteristics for training
    UIEB and EUVP are used for training to learn halo laws.

pith-pipeline@v0.9.0 · 5542 in / 1303 out tokens · 47233 ms · 2026-05-12T04:29:24.091657+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    Computer modeling and the design of optimal un- derwater imaging systems,

    J. Jaffe, “Computer modeling and the design of optimal un- derwater imaging systems,”IEEE Journal of Oceanic En- gineering, vol. 15, no. 2, pp. 101–111, 1990

    work page 1990

  2. [2]

    Visual in-context learning for underwa- ter image restoration,

    G. Jiang et al., “Visual in-context learning for underwa- ter image restoration,”IEEE Signal ProcessingLetters, vol. 33, pp. 1072–1076, 2026

    work page 2026

  3. [3]

    A lightweight polarization-guided plug-in for underwater image enhancement,

    G. Ju et al., “A lightweight polarization-guided plug-in for underwater image enhancement,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 36, no. 3, pp. 2729–2743, 2026

    work page 2026

  4. [4]

    Multi-parameter detection based on u- shaped biomimetic optical fiber sensor,

    Y . Wang et al., “Multi-parameter detection based on u- shaped biomimetic optical fiber sensor,”Journal of Light- wave Technology, vol. 43, no. 16, pp. 7954–7963, 2025

    work page 2025

  5. [5]

    Vignetting correction using an optical model and constant chromaticity prior,

    B. Yu et al., “Vignetting correction using an optical model and constant chromaticity prior,”IEEE Transactions on Computational Imaging, vol. 9, pp. 1071–1083, 2023

    work page 2023

  6. [6]

    Underwater depth estimation and im- age restoration based on single images,

    P. L. Drews et al., “Underwater depth estimation and im- age restoration based on single images,”IEEE Computer Graphics and Applications, vol. 36, no. 2, pp. 24–35, 2016

    work page 2016

  7. [7]

    Tebcf: Real-world underwater image tex- ture enhancement model based on blurriness and color fu- sion,

    J. Yuan et al., “Tebcf: Real-world underwater image tex- ture enhancement model based on blurriness and color fu- sion,”IEEE Transactions on Geoscience and Remote Sens- ing, vol. 60, pp. 1–15, 2022. 8

    work page 2022

  8. [8]

    Unsupervised underwater image restoration: From a homology perspective,

    Z. Fu et al., “Unsupervised underwater image restoration: From a homology perspective,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 1, pp. 643–651, Jun. 2022

    work page 2022

  9. [9]

    Non-uniform illumination underwater im- age restoration via illumination channel sparsity prior,

    G. Hou et al., “Non-uniform illumination underwater im- age restoration via illumination channel sparsity prior,” IEEE Transactions on Circuits and Systems for Video Tech- nology, vol. 34, no. 2, pp. 799–814, 2024

    work page 2024

  10. [10]

    Low-light enhancement using a plug-and-play retinex model with shrinkage mapping for illumination estimation,

    Y .-H. Lin and Y .-C. Lu, “Low-light enhancement using a plug-and-play retinex model with shrinkage mapping for illumination estimation,”IEEE Transactions on Image Pro- cessing, vol. PP, pp. 1–1, 07 2022

    work page 2022

  11. [11]

    Color balance and fusion for under- water image enhancement,

    C. O. Ancuti et al., “Color balance and fusion for under- water image enhancement,”IEEE Transactions on Image Processing, vol. 27, no. 1, pp. 379–393, 2018

    work page 2018

  12. [12]

    Underwater image enhancement with hyper-laplacian reflectance priors,

    P. Zhuang et al., “Underwater image enhancement with hyper-laplacian reflectance priors,”IEEE Transactions on Image Processing, vol. 31, pp. 5442–5455, 2022

    work page 2022

  13. [13]

    A vignetting-correction-based underwater image enhancement method for auv with artificial light,

    Z. Mi et al., “A vignetting-correction-based underwater image enhancement method for auv with artificial light,” IEEE Journal of Oceanic Engineering, vol. 50, no. 1, pp. 213–227, 2025

    work page 2025

  14. [14]

    Underwater image enhancement via weighted wavelet visual perception fusion,

    W. Zhang et al., “Underwater image enhancement via weighted wavelet visual perception fusion,”IEEE Trans- actions on Circuits and Systems for Video Technology, vol. 34, no. 4, pp. 2469–2483, 2024

    work page 2024

  15. [15]

    Mda-net: A multi-distribution aware net- work for underwater image enhancement,

    X. Liu et al., “Mda-net: A multi-distribution aware net- work for underwater image enhancement,”IEEE Transac- tions on Geoscience and Remote Sensing, vol. 63, pp. 1–13, 2025

    work page 2025

  16. [16]

    Fast underwater image enhancement for improved visual perception,

    M. J. Islam et al., “Fast underwater image enhancement for improved visual perception,”IEEE Robotics and Automa- tion Letters, vol. 5, no. 2, pp. 3227–3234, 2020

    work page 2020

  17. [17]

    Underwater scene prior inspired deep under- water image and video enhancement,

    C. Li et al., “Underwater scene prior inspired deep under- water image and video enhancement,”Pattern Recognition, vol. 98, p. 107038, 2020

    work page 2020

  18. [18]

    An underwater image enhancement bench- mark dataset and beyond,

    C. Li et al., “An underwater image enhancement bench- mark dataset and beyond,”IEEE Transactions on Image Processing, vol. 29, pp. 4376–4389, 2020

    work page 2020

  19. [19]

    U-shape transformer for underwater im- age enhancement,

    L. Peng et al., “U-shape transformer for underwater im- age enhancement,”IEEE Transactions on Image Process- ing, vol. 32, pp. 3066–3079, 2023

    work page 2023

  20. [20]

    Underwater image enhancement via brightness mask-guided multi-attention embedding,

    Y . Li et al., “Underwater image enhancement via brightness mask-guided multi-attention embedding,”Signal Process- ing: Image Communication, vol. 130, p. 117200, 2025

    work page 2025

  21. [21]

    L2uwe: A frame- work for the efficient enhancement of low-light underwa- ter images using local contrast and multi-scale fusion,

    T. P. Marques and A. Branzan Albu, “L2uwe: A frame- work for the efficient enhancement of low-light underwa- ter images using local contrast and multi-scale fusion,” in 2020IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW), 2020, pp. 2286– 2295

    work page 2020

  22. [22]

    Shallow-uwnet: Compressed model for un- derwater image enhancement (student abstract),

    A. Naik et al., “Shallow-uwnet: Compressed model for un- derwater image enhancement (student abstract),”Proceed- ings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 18, pp. 15 853–15 854, May 2021

    work page 2021

  23. [23]

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

    S. Huang et al., “Contrastive semi-supervised learning for underwater image restoration via reliable bank,” in2023 IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR), 2023, pp. 18 145–18 155

    work page 2023

  24. [24]

    Deep color-corrected multiscale retinex net- work for underwater image enhancement,

    H. Qi et al., “Deep color-corrected multiscale retinex net- work for underwater image enhancement,”IEEE Transac- tions on Geoscience and Remote Sensing, vol. 62, pp. 1–13, 2024

    work page 2024

  25. [25]

    Parameter-free multi-view clustering via refined tensor learning,

    J. Yang et al., “Parameter-free multi-view clustering via refined tensor learning,”Neurocomputing, vol. 656, p. 131497, 2025

    work page 2025

  26. [26]

    Bi-level inter-modality modulation for un- supervised visible-infrared person re-identification,

    J. Peng et al., “Bi-level inter-modality modulation for un- supervised visible-infrared person re-identification,”IEEE Transactions on Information Forensics and Security, vol. PP, pp. 1–1, 01 2026

    work page 2026

  27. [27]

    Tensor completion framework by graph refinement for incomplete multi-view clustering,

    H. Wang et al., “Tensor completion framework by graph refinement for incomplete multi-view clustering,”IEEE Transactions on Multimedia, vol. 27, pp. 9385–9398, 2025

    work page 2025

  28. [28]

    Graph-collaborated auto-encoder hash- ing for multiview binary clustering,

    H. Wang et al., “Graph-collaborated auto-encoder hash- ing for multiview binary clustering,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 7, pp. 10 121–10 133, 2024

    work page 2024

  29. [29]

    Amplitude - phase decomposition-based latent diffusion model for underwater image enhance- ment,

    H. Zhang et al., “Amplitude - phase decomposition-based latent diffusion model for underwater image enhance- ment,”Expert Systems with Applications, vol. 312, p. 131296, 2026

    work page 2026

  30. [30]

    Adaptive memorization with group labels for unsupervised person re-identification,

    J. Peng et al., “Adaptive memorization with group labels for unsupervised person re-identification,”IEEE Transac- tions on Circuits and Systems for Video Technology, vol. 33, no. 10, p. 5802–5813, Oct. 2023

    work page 2023

  31. [31]

    Between/within view information complet- ing for tensorial incomplete multi-view clustering,

    M. Yao et al., “Between/within view information complet- ing for tensorial incomplete multi-view clustering,”IEEE Transactions on Multimedia, vol. 27, pp. 1538–1550, 2025

    work page 2025

  32. [32]

    Consensus-guided incomplete multi-view clustering via cross-view affinities learning,

    Q. Liu et al., “Consensus-guided incomplete multi-view clustering via cross-view affinities learning,” inProceed- ings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25, International Joint Con- ferences on Artificial Intelligence Organization, 8 2025, pp. 5761–5769, main Track

    work page 2025

  33. [33]

    Focus more on what? guiding multi-task training for end-to-end person search,

    B. Cai et al., “Focus more on what? guiding multi-task training for end-to-end person search,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 7, pp. 7266– 7278, 2025

    work page 2025

  34. [34]

    Manifold-based incomplete multi-view clustering via bi-consistency guidance,

    H. Wang et al., “Manifold-based incomplete multi-view clustering via bi-consistency guidance,”IEEE Transac- tions on Multimedia, vol. 26, pp. 10 001– 10 014, 2024

    work page 2024

  35. [35]

    Omni contextual aggregation networks for high-fidelity image inpainting,

    J. Peng et al. “Omni contextual aggregation networks for high-fidelity image inpainting,”IEEE Transactions on Cir- cuits and Systems for Video Technology, vol. 35, no. 6, pp. 6129– 6144, 2025

    work page 2025

  36. [36]

    Reliable feature imputation with cross- view relation transfer for deep incomplete multi-view clas- sification,

    G. Jiang et al., “Reliable feature imputation with cross- view relation transfer for deep incomplete multi-view clas- sification,”IEEE Transactions on Circuits and Systems for Video Technology, pp. 1–1, 2026

    work page 2026

  37. [37]

    Amplitude exchanging network for unsu- pervised underwater image enhancement,

    R. Zeng et al., “Amplitude exchanging network for unsu- pervised underwater image enhancement,”Pattern Recog- nition, vol. 175, p. 113119, 2026. 9

    work page 2026

  38. [38]

    Dcd-uie: Decoupled chromatic diffusion model for underwater image enhancement,

    G. Fan et al., “Dcd-uie: Decoupled chromatic diffusion model for underwater image enhancement,”IEEE Trans- actions on Image Processing, vol. 35, pp. 449–464, 2026

    work page 2026

  39. [39]

    Refid: Reciprocal frequency- aware gener- alizable person re-identification via decomposition and fil- tering,

    J. Peng et al., “Refid: Reciprocal frequency- aware gener- alizable person re-identification via decomposition and fil- tering,”ACM Trans. Multimedia Comput. Commun. Appl., vol. 20, no. 7, Apr. 2024

    work page 2024

  40. [40]

    Hierarchical sequential context modelling for high-fidelity image inpainting,

    Z. Sun et al., “Hierarchical sequential context modelling for high-fidelity image inpainting,”IEEE Transactions on Circuits and Systems for Video Technology, pp. 1–1, 2025

    work page 2025

  41. [41]

    Hybrid anchor graph learning and ten- sorized spectral embedding fusion for multi-view cluster- ing,

    G. Jiang et al., “Hybrid anchor graph learning and ten- sorized spectral embedding fusion for multi-view cluster- ing,”Neurocomputing, vol. 673, p. 132768, 2026

    work page 2026

  42. [42]

    Hierarchical structure-guided incomplete multi-view tensor clustering,

    J. Yang et al., “Hierarchical structure-guided incomplete multi-view tensor clustering,”Applied Intelligence, vol. 56, no. 4, Feb. 2026

    work page 2026

  43. [43]

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

    H. Wang et al., “Tensorized parameter-free multi-view spectral clustering based on fair representation learning,” IEEE Transactions on Multimedia, pp. 1–13, 2026

    work page 2026

  44. [44]

    Unsupervised lifelong person re- identification via affinity harmonization,

    J. Tan et al., “Unsupervised lifelong person re- identification via affinity harmonization,”ACM Trans. Multimedia Comput. Commun. Appl., vol. 22, no. 4, Mar. 2026

    work page 2026

  45. [45]

    Improving underwater visibility us- ing vignetting correction,

    K. Sooknanan et al., “Improving underwater visibility us- ing vignetting correction,” inVisual Information Process- ing and Communication III, vol. 8305, International Soci- ety for Optics and Photonics. SPIE, 2012, p. 83050M

    work page 2012

  46. [46]

    Underwater image devignetting and colour correction,

    Y . Li et al., “Underwater image devignetting and colour correction,” inImage and Graphics. Cham: Springer Inter- national Publishing, 2015, pp. 510–521

    work page 2015

  47. [47]

    Watergan: Unsupervised generative network to enable real-time color correction of monocular underwater images,

    J. Li et al., “Watergan: Unsupervised generative network to enable real-time color correction of monocular underwater images,”IEEE Robotics and Automation Letters, vol. PP, 02 2017

    work page 2017

  48. [48]

    Underwater vignetting image correc- tion based on binary polynomial regularization and la- tent low-rank representation,

    Y . Wang et al., “Underwater vignetting image correc- tion based on binary polynomial regularization and la- tent low-rank representation,”IEEE Transactions on Cir- cuits and Systems for Video Technology, vol. 35, no. 4, pp. 3410–3425, 2025. 10

    work page 2025