Pith. sign in

REVIEW 4 major objections 4 minor 61 references

UniDemoir\'e: Towards Universal Image Demoir\'eing with Data Generation and Synthesis

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

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

Pith's one-line read The paper claims that synthetic moiré data, generated by separating patterns from content and refining their tone, can train demoiréing models that generalize to unseen moiré domains.

desk verdict A credible data-generation pipeline with consistent zero-shot gains, but the 'universal' claim outruns the single-run, three-dataset evidence. read the letter →

arxiv 2502.06324 v1 pith:ADKLQJXK submitted 2025-02-10 cs.CV cs.AI

Zemin Yang , Yujing Sun , Xidong Peng , Siu Ming Yiu , Yuexin Ma This is my paper · ORCID
classification cs.CVcs.AI
keywords imagedemoiréingmoirépatternsynthesisdatagenerationlatentdiffusionmodeltonerefinementnetworkcross-domaingeneralizationzero-shotsynthetictraining
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

UniDemoiré attacks the data bottleneck in image demoiréing by manufacturing training pairs instead of collecting them. The paper argues that moiré patterns are independent of the image content they overlay, so patterns filmed against a plain white screen can be freely composited onto clean natural images. A latent-diffusion generator expands the captured pattern set, and a learnable tone-refinement network makes the composites match the color and brightness of real screen-capture moiré. If the approach is right, demoiréing models trained on this synthetic data should transfer to moiré domains they have never seen, and the paper reports that they do, beating earlier synthesis pipelines in zero-shot and cross-dataset tests. The central bet is that data-engineered diversity plus realism is enough to build a universal demoiréing model.

What carries the argument

The load-bearing mechanism is the content-independence decomposition of a moiré-captured image into a foreground moiré pattern layer and a background content layer. Carrying the argument are three components built on that decomposition: a real-captured 4K moiré pattern dataset of 150,000 patterns with diversity in zoom rate, CMOS, and panel type; a latent diffusion model that generates additional patterns from a compressed representation of the real ones; and a two-step synthesis stage in which Multiply and Grain Merge blending is followed by a Tone Refinement Network, a U-shaped transformer whose training-only feature-statistics fusion block mixes tone statistics between the synthesized and real images. The refinement network, trained by perceptual, RGB-uv histogram color, and total-variation losses, is what closes the real-to-synthetic gap in tone, so the synthesized data behave like real screen captures.

What would settle it

A direct test would photograph the same moiré-inducing setup twice, once with a plain white screen and once with a high-frequency natural texture displayed behind a fixed pattern, and compare the extracted pattern layers pixel-wise; if the pattern layer changes measurably with the underlying content instead of staying constant, the white-screen capture premise is wrong.

Watch

Extended reading notes

Core claim

The paper's central claim is that a universal image demoiréing model can be trained on purely synthesized data, provided the synthesis separates the moiré pattern from the image content and then restores realism. The discovery is that this separation works: 150,000 real moiré patterns captured at 4K against white backgrounds, augmented by latent-diffusion sampling, can be blended with clean natural images via multiply-plus-grain-merge composition and a tone-refinement transformer that mimics the color and brightness statistics of real moiré images. Across zero-shot evaluations on real benchmarks, the resulting training data yields demoiréing networks that outperform networks trained with prior synthetic pipelines, and adding the synthesized images to real training data improves cross-dataset transfer on every source-target pair tested.

Load-bearing premise

The approach stands or falls on the premise that a moiré pattern does not depend on the image content it appears over, so patterns filmed against a blank white screen and composited onto natural images, after tone refinement, represent real on-screen moiré faithfully.

Editorial extensions

If this is right

  • Zero-shot transfer becomes practical: a demoiréing network trained only on synthesized images can be deployed on real screen-capture benchmarks it never saw, with the paper reporting the largest gains on the most difficult 4K dataset.
  • Real training data becomes optional for the demoiréing backbone: synthesis supplies the volume and diversity, while the tone-refinement network, which does use real moiré images as guidance, carries the realism.
  • One clean image can be paired with many different moiré patterns, multiplying the training set and breaking the one-to-one clean-to-moiré alignment bottleneck that limits current datasets.
  • Because patterns are stored separately from content, the same pipeline scales to arbitrarily many composite images and, as the paper argues, to much larger demoiréing model capacities.

Reading between the lines

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

  • If the content-independence premise holds, the same white-screen capture plus tone-refinement recipe could be applied to other screen-capture artifacts such as specular reflections, glare, or color cast, where the artifact is multiplicative or additive over content.
  • The evaluation leaves open whether the observed generalization comes mainly from pattern diversity, via the latent-diffusion generator, or from tone realism, via the refinement network; an ablation that varies one while holding the other fixed would separate the two sources of gain.
  • The tone-refinement network is itself trained on real moiré images from standard benchmarks, so the 'synthetic data only' claim applies to the demoiréing backbone, not the whole pipeline; a stricter test would train the refinement network on a disjoint set of real moiré images and then evaluate the final model on unseen domains.
  • A fairer universal-generalization benchmark would hold out target panel types, zoom settings, or phone sensors from both the pattern dataset and the refinement-network training set, since current cross-dataset tests share the general screen-photo domain with the benchmarks that guided tone refinement.
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 / 4 minor

Summary. The paper proposes UniDemoiré, a data-generation pipeline for image demoiréing. It collects a 150k-pattern dataset of moiré artifacts captured on plain white screens using six phones and six displays, trains a latent diffusion model to generate additional patterns after multi-scale cropping and sharpness/colorfulness filtering, and synthesizes training images by blending patterns with clean natural images using multiply and grain-merge operations followed by a Uformer-based tone refinement network trained with perceptual, histogram-color, and TV losses. The authors train MBCNN and ESDNet-L on the synthesized data and report zero-shot and cross-dataset results on TIP, FHDMi, and UHDM, with ablations showing that each component contributes.

Significance. If the results hold, the paper offers a practical recipe for expanding demoiréing training data without costly aligned capture. Its strengths are the release of code, comparison against three external synthesis baselines (Shooting, UnDeM, MoireSpace), and the consistent direction of the quantitative gains. However, the significance is tempered by the narrow evaluation regime (three screen-photo datasets whose capture conditions overlap with the collected data), the single-run reporting with no error bars, and the untested content-independence assumption underlying the synthesis. The headline 'universal' claim therefore currently exceeds the demonstrated scope, and the 150k-pattern dataset is not released, limiting reproducibility of the data-generation core.

major comments (4)
  1. [Abstract; Appendix C.5] The central 'universal image demoiréing solution' claim is stronger than the experimental evidence. All evaluations use only TIP, FHDMi, and UHDM, which are screen-photo datasets captured under conditions that overlap substantially with the authors' own collection (six mobile phones and six IPS/SVA screens in Table 1 and Appendix A.2). Appendix C.5 itself documents a failure case where target-domain moiré differs strongly from the source domain. The claim should either be restricted to the screen-photo regime or supported by out-of-domain evaluations (e.g., different panel technologies beyond IPS/SVA, printed materials, or other optical setups).
  2. [Tables 2, 3; Appendix C.1] All quantitative results come from a single training run with the seed fixed to values from prior work, as stated in Appendix C.1. Several cross-dataset gains over the strongest baseline are small: for example, Table 3 shows UHDM→FHDMi SSIM 0.7525 versus 0.7496 with MBCNN, and TIP→FHDMi LPIPS 0.2315 versus 0.2382 with ESDNet-L. Without repeated runs, error bars, or significance tests, the direction of these differences could be within seed-to-seed variation. Please report mean±std over multiple seeds or paired significance tests for the main zero-shot and cross-dataset tables, and consider ablating on more than the single UHDM→FHDMi configuration used in Tables 4 and 8.
  3. [Introduction; Eq. (1); Appendix C.5] The content-independence premise — that moiré patterns captured against a plain white screen can be composited onto arbitrary natural images and remain representative — is load-bearing but not directly tested. The introduction states this premise and Eq. (1) implements it as multiplicative blending, but the paper provides no quantitative comparison between synthesized moiré images and real moiré images with identical scene content, nor any content-stratified analysis. The failure case in Appendix C.5 suggests that pattern-content interactions can matter. A direct validation (e.g., distribution-distance statistics between synthetic and real moiré images, or a user study comparing realism) would substantially strengthen the generalization claim.
  4. [Table 2; Section B.2] The zero-shot comparison is not fully symmetric in how real data are used. In the paragraph before Table 2, the authors state that real moiré images from TIP are used to train their own TRN, while UnDeM uses TIP real moiré images for training and inference under the † variant. The demoiréing networks are trained only on synthesized data in both cases, but the synthesis modules receive different amounts of real-data supervision. Please make this distinction explicit in the main text and discuss whether the additional real-data guidance explains part of the performance gap.
minor comments (4)
  1. [Section B.1] In the latent diffusion implementation, the paper states that a downsampling factor of f=32 and 64 hidden channels give a latent variable z of dimension 64×64×24; for 768×768 input patches, the spatial dimension should be 24×24, so the reported dimension appears to be a typo (likely 24×24×64).
  2. [Equation (5)] Equation (5) defines r_x using op_x and op_n, but the main text never defines op_n or states the ranges of op_m and op_g; the values appear only in Appendix B.2. Please move or repeat these definitions in the main text.
  3. [Table 1] In Table 1, the panel column reads 'IPS, SV A'; this should be 'IPS, SVA' with a definition of SVA, and the table header 'Moiré Image Dataset' is difficult to parse due to line breaks.
  4. [Related Work] There are several typos and notation inconsistencies: 'demoreing' in the Related Work section, inconsistent use of 'MoireSpace' versus 'MoiréSpace', and the garbled sentence 'The flow of I... in the TRN' in the Figure 4 caption. A thorough proofread is needed.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the only self-citation (MoireSpace) is used as a baseline and inspiration, not as load-bearing support for the central claim.

full rationale

The paper's central claim is an empirical data-generation pipeline whose output is benchmarked against external methods (Shooting, UnDeM, MoireSpace) and real datasets (TIP, FHDMi, UHDM). No fitted parameter is renamed as a prediction: the Tone Refinement Network is trained on real source-domain moiré images and used to synthesize training images, and the downstream demoiréing models are evaluated on held-out target domains. The content-independence premise is a stated physical assumption, not an equation-level self-reference. The only self-citation is MoireSpace (Yang et al. 2023), which shares an author with this paper; it is used as inspiration for the pattern-capture idea and as a baseline that the paper critiques rather than as the justification for the claimed generalization. Appendix C.5 concedes failure cases for very different target domains, which is a scope limitation, not circularity. Therefore no circular step exists; the score of 2 reflects only the minor non-load-bearing self-citation.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the assumption that moiré patterns can be decoupled from image content and re-composited, and that learned tone refinement transfers across domains. No new physical entities are introduced. Free parameters are standard hand-set hyperparameters of the data pipeline, not fitted scientific constants.

free parameters (5)
  • sharpness threshold delta_s = 15
    Hand-picked threshold in Moiré Pattern Generation data filtering (Algorithm 1); affects which patches train the diffusion model.
  • colorfulness threshold delta_c = 2
    Hand-picked threshold in the same filtering; controls diversity of training patches.
  • blending weights omega_m, omega_g = omega_m in [0.65, 0.75], omega_g = 1 - omega_m
    Hand-set composition ratios in Moiré Image Blending (Eq. 6), ablated in Appendix C.4.
  • opacities op_m, op_g = 1.0, 0.8
    Hand-set opacities for Multiply and Grain Merge layers (Eq. 5).
  • loss weights lambda_per, lambda_color, lambda_tv = 1.0, 1.0, 0.1
    Weighted compound loss in Eq. 13; set by hand.
assumptions (3)
  • domain assumption Moiré pattern is independent of image content
    Stated in the Introduction ('inspired by the fact that the moiré pattern is unrelated to the content of the image') and used to justify capturing patterns on a white background and compositing them onto arbitrary natural images.
  • domain assumption A tone refinement network trained on real moiré images (TIP/FHDMi/UHDM) transfers to other target domains
    TRN is trained on specific real datasets and then used to synthesize training data for zero-shot or cross-dataset evaluation on other domains.
  • domain assumption Latent diffusion can generate novel, realistic moiré patterns from the collected pattern distribution
    The MPG assumes that sampling from the learned latent distribution produces patterns that are as useful as real captures for downstream training.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UniDemoir\'e: Towards Universal Image Demoir\'eing with Data Generation and Synthesis." pith.science (2026). https://pith.science/paper/ADKLQJXK

@misc{pith2026250206324,
  author       = {Pith},
  title        = {Pith review of: UniDemoir\'e: Towards Universal Image Demoir\'eing with Data Generation and Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ADKLQJXK}},
  note         = {Machine review of arXiv:2502.06324}
}
read the original abstract

Image demoir\'eing poses one of the most formidable challenges in image restoration, primarily due to the unpredictable and anisotropic nature of moir\'e patterns. Limited by the quantity and diversity of training data, current methods tend to overfit to a single moir\'e domain, resulting in performance degradation for new domains and restricting their robustness in real-world applications. In this paper, we propose a universal image demoir\'eing solution, UniDemoir\'e, which has superior generalization capability. Notably, we propose innovative and effective data generation and synthesis methods that can automatically provide vast high-quality moir\'e images to train a universal demoir\'eing model. Our extensive experiments demonstrate the cutting-edge performance and broad potential of our approach for generalized image demoir\'eing.

Figures

Figures reproduced from arXiv: 2502.06324 by the authors.

Figure 1
Figure 1. The workflow of our proposed UniDemoire.´ panel types. Second, building on this real-captured moire´ pattern dataset, we propose a diffusion model-based Moire´ Pattern Generation method to further increase the diver￾sity of moire patterns. Specifically, we implement a multi- ´ scale cropping strategy to accommodate different input im￾age sizes and an effective data filtering strategy to ensure the quality of trainin… view at source ↗
Figure 2
Figure 2. Data collection setup (left), and examples of moir [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. lower-right, an increased sharpness value indicates more visible moire patterns, while an increased colorfulness ´ value signifies patterns with richer colors. The sharpness metric is calculated as the standard deviation of grayscaled input image processed with an edge filter, while the colorful￾ness metric is calculated as the average standard deviations of A and B channels in image LAB color space. Learning Moire … view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Overview of the Moire Image Synthesis stage (a). It involves a Moir ´ e Image Blending module (b) for initial moir ´ e´ image synthesis and a Tone Refinement Network (c) to refine for more realistic results. The Tone Refinement Network (TRN) proposed here is built on a…
Figure 5
Figure 5. Figure 5: Visualization of our intermediate synthetic results. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Comparisons of demoireing results. ´ implicit moire synthesis approach “UnDeM” (Zhong et al. ´ 2024), which employs a neural network, and the explicit synthesis method termed “MoireSpace” (Yang et al. 2023), which utilizes its moire pattern dataset. ´ Demoireing Models…
Figure 7
Figure 7. Figure 7: Comparison of sharpness and colorfulness be [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Samples from MoireSpace (Yang et al. 2023) and our 4K Moir [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Visualization of sampled patches using our Moir [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Visualization of our intermediate synthetic results. The final synthesis of [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Examples of the “Checkerboard Artifacts” that occur in the [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Examples of the MHRNID dataset. where k(·) is the inverse-quadratic kernel: k (Iuc, Ivc, u, v) = (1 + (|Iuc − u| /τ ) 2 −1 ×  1 + (|Ivc − v| /τ ) 2 −1 (21) where τ is a fall-off parameter to control the smoothness of the histogram’s bins. Finally, the histogram fea…
Figure 13
Figure 13. Figure 13: Failure Examples. et al. 2022). Our method’s model effectively removes moire´ artifacts and retains high-frequency details, indicating the strong generalization ability of our proposed UniDemoire.´ C.3 Runtime Comparisons [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: Qualitative comparisons of synthesized moire images were obtained using the shooting method, UnDeM, MoireS [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Qualitative comparisons of zero-shot evaluation on the UHDM dataset. [PITH_FULL_IMAGE:figures/full_fig_p021_15.png]
Figure 16
Figure 16. Figure 16: Qualitative comparisons of our models with other state-of-the-art methods on the FHDMi dataset. [PITH_FULL_IMAGE:figures/full_fig_p022_16.png]
Figure 17
Figure 17. Figure 17: Qualitative comparisons of our models with other state-of-the-art methods on the TIP dataset. [PITH_FULL_IMAGE:figures/full_fig_p023_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 40 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Afifi, M.; and Brown, M. S. 2019. Sensor-Independent Illumination Estimation for DNN Models. arXiv:1912.06888

    arXiv 2019

  4. [4]

    A.; and Brown, M

    Afifi, M.; Brubaker, M. A.; and Brown, M. S. 2021. HistoGAN: Controlling Colors of GAN-Generated and Real Images via Color Histograms. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    work page 2021

  5. [5]

    Afifi, M.; Price, B.; Cohen, S.; and Brown, M. S. 2019. When Color Constancy Goes Wrong: Correcting Improperly White-Balanced Images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    work page 2019

  6. [6]

    Amidror, I. 2009. The Theory of the Moir \'e Phenomenon: Volume I: Periodic Layers , volume 38. Springer Science & Business Media

    work page 2009

  7. [7]

    Barron, J. T. 2015. Convolutional Color Constancy. arXiv:1507.00410

    work page Pith review arXiv 2015

  8. [8]

    Dhariwal, P.; and Nichol, A. 2021. Diffusion Models Beat GANs on Image Synthesis. arXiv:2105.05233

    arXiv 2021

Show all 61 references
  1. [9]

    Dosovitskiy, A.; and Brox, T. 2016. Generating Images with Perceptual Similarity Metrics based on Deep Networks. Neural Information Processing Systems,Neural Information Processing Systems

    2016

  2. [10]

    Eibenberger, E.; and Angelopoulou, E. 2012. The importance of the normalizing channel in log-chromaticity space. In 2012 19th IEEE International Conference on Image Processing

    2012

  3. [11]

    Esser, P.; Rombach, R.; and Ommer, B. 2021. Taming Transformers for High-Resolution Image Synthesis. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2021

  4. [12]

    Fei, B.; Lyu, Z.; Pan, L.; Zhang, J.; Yang, W.; Luo, T.; Zhang, B.; and Dai, B. 2023. Generative Diffusion Prior for Unified Image Restoration and Enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9935--9946

    2023

  5. [13]

    GIMP. 2023. Layer Blending Modes. [Online; accessed 10-April-2024]

    2023

  6. [14]

    He, B.; Wang, C.; Shi, B.; and Duan, L.-Y. 2019. Mop moire patterns using mopnet. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 2424--2432

    2019

  7. [15]

    He, B.; Wang, C.; Shi, B.; and Duan, L.-Y. 2020. FHDe 2 Net: Full High Definition Demoireing Network. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXII 16, 713--729. Springer

    2020

  8. [16]

    Hosu, V.; Agnolucci, L.; Wiedemann, O.; and Iso, D. 2024. UHD-IQA Benchmark Database: Pushing the Boundaries of Blind Photo Quality Assessment. arXiv:2406.17472

    2024 arXiv

  9. [17]

    Huang, X.; and Belongie, S. 2017. Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization. In 2017 IEEE International Conference on Computer Vision (ICCV)

    2017

  10. [18]

    Johnson, J.; Alahi, A.; and Fei-Fei, L. 2016. Perceptual losses for real-time style transfer and super-resolution. In European conference on computer vision, 694--711. Springer

    2016

  11. [19]

    P.; and Ba, J

    Kingma, D. P.; and Ba, J. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980

    2014 arXiv

  12. [20]

    Lee, J.; Son, H.; Rim, J.; Cho, S.; and Lee, S. 2021. Iterative Filter Adaptive Network for Single Image Defocus Deblurring. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2034--2042

    2021

  13. [21]

    Li, H.; Li, J.; Zhao, D.; and Xu, L. 2021. Dehazeflow: Multi-scale conditional flow network for single image dehazing. In Proceedings of the 29th ACM International Conference on Multimedia, 2577--2585

    2021

  14. [22]

    Liu, F.; Yang, J.; and Yue, H. 2015. Moir \'e pattern removal from texture images via low-rank and sparse matrix decomposition. In 2015 Visual Communications and Image Processing (VCIP), 1--4. IEEE

    2015

  15. [23]

    Liu, L.; An, J.; Yuan, S.; Zhou, W.; Li, H.; Wang, Y.; and Tian, Q. 2024. Video Demoir \'e ing with Deep Temporal Color Embedding and Video-Image Invertible Consistency. IEEE Transactions on Multimedia

    2024

  16. [24]

    Liu, L.; Liu, J.; Yuan, S.; Slabaugh, G.; Leonardis, A.; Zhou, W.; and Tian, Q. 2020. Wavelet-based dual-branch network for image demoir \'e ing. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XIII 16, 86--102. Springer

    2020

  17. [25]

    Loshchilov, I.; and Hutter, F. 2016. Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983

    2016 arXiv

  18. [26]

    Loshchilov, I.; and Hutter, F. 2019. Decoupled Weight Decay Regularization. arXiv:1711.05101

    2019 arXiv

  19. [27]

    Luo, X.; Zhang, J.; Hong, M.; Qu, Y.; Xie, Y.; and Li, C. 2020. Deep wavelet network with domain adaptation for single image demoireing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 420--421

    2020

  20. [28]

    o lund, J.; and Sch \

    Luo, Z.; Gustafsson, F. K.; Zhao, Z.; Sj \"o lund, J.; and Sch \"o n, T. B. 2023. Refusion: Enabling large-size realistic image restoration with latent-space diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1680--1691

    2023

  21. [29]

    Niu, D.; Guo, R.; and Wang, Y. 2021. Mori \'e attack (ma): A new potential risk of screen photos. Advances in Neural Information Processing Systems, 34: 26117--26129

    2021

  22. [30]

    Niu, Y.; Lin, Z.; Liu, W.; and Guo, W. 2023. Progressive Moire Removal and Texture Complementation for Image Demoireing. IEEE Transactions on Circuits and Systems for Video Technology

    2023

  23. [31]

    Odena, A.; Dumoulin, V.; and Olah, C. 2016. Deconvolution and Checkerboard Artifacts. Distill

    2016

  24. [32]

    G.; Kim, H.; Koh, Y

    Park, H.; Vien, A. G.; Kim, H.; Koh, Y. J.; and Lee, C. 2022. Unpaired screen-shot image demoir \'e ing with cyclic moir \'e learning. IEEE Access, 10: 16254--16268

    2022

  25. [33]

    Porter, T.; and Duff, T. 1984. Compositing digital images. ACM SIGGRAPH Computer Graphics, 253–259

    1984

  26. [34]

    Rombach, R.; Blattmann, A.; Lorenz, D.; Esser, P.; and Ommer, B. 2022. High-Resolution Image Synthesis with Latent Diffusion Models. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  27. [35]

    Ronneberger, O.; Fischer, P.; and Brox, T. 2015. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, 234--241. Springer

    2015

  28. [36]

    Simonyan, K.; and Zisserman, A. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556

    2014 arXiv

  29. [37]

    Song, J.; Meng, C.; and Ermon, S. 2022. Denoising Diffusion Implicit Models. arXiv:2010.02502

    2022 arXiv

  30. [38]

    Sun, B.; Li, S.; and Sun, J. 2014. Scanned image descreening with image redundancy and adaptive filtering. IEEE transactions on image processing, 23(8): 3698--3710

    2014

  31. [39]

    Sun, Y.; Yu, Y.; and Wang, W. 2018. Moir \'e photo restoration using multiresolution convolutional neural networks. IEEE Transactions on Image Processing, 27(8): 4160--4172

    2018

  32. [40]

    Ulyanov, D.; Vedaldi, A.; and Lempitsky, V. 2016. Instance Normalization: The Missing Ingredient for Fast Stylization. arXiv: Computer Vision and Pattern Recognition,arXiv: Computer Vision and Pattern Recognition

    2016

  33. [41]

    Wang, C.; He, B.; Wu, S.; Wan, R.; Shi, B.; and Duan, L.-Y. 2023 a . Coarse-to-fine Disentangling Demoir \'e ing Framework for Recaptured Screen Images. IEEE Transactions on Pattern Analysis and Machine Intelligence

    2023

  34. [42]

    C.; and Lin, D

    Wang, J.; Chen, K.; Xu, R.; Liu, Z.; Loy, C. C.; and Lin, D. 2019. CARAFE: Content-Aware ReAssembly of FEatures. arXiv:1905.02188

    2019 arXiv

  35. [43]

    Wang, T.; Zhang, K.; Shen, T.; Luo, W.; Stenger, B.; and Lu, T. 2023 b . Ultra-high-definition low-light image enhancement: A benchmark and transformer-based method. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, 2654--2662

    2023

  36. [44]

    Wang, X.; Xie, L.; Dong, C.; and Shan, Y. 2021. Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 1905--1914

    2021

  37. [45]

    C.; Sheikh, H

    Wang, Z.; Bovik, A. C.; Sheikh, H. R.; and Simoncelli, E. P. 2004. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4): 600--612

    2004

  38. [46]

    Wang, Z.; Cun, X.; Bao, J.; Zhou, W.; Liu, J.; and Li, H. 2022. Uformer: A General U-Shaped Transformer for Image Restoration. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    2022

  39. [47]

    Xing, W.; and Egiazarian, K. 2021. End-to-End Learning for Joint Image Demosaicing, Denoising and Super-Resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 3507--3516

    2021

  40. [48]

    Yang, C.; Yang, Z.; Ke, Y.; Chen, T.; Grzegorzek, M.; and See, J. 2023. Doing More With Moir \'e Pattern Detection in Digital Photos. IEEE Transactions on Image Processing, 32: 694--708

    2023

  41. [49]

    Yang, J.; Liu, F.; Yue, H.; Fu, X.; Hou, C.; and Wu, F. 2017 a . Textured image demoir \'e ing via signal decomposition and guided filtering. IEEE Transactions on Image Processing, 26(7): 3528--3541

    2017

  42. [50]

    Yang, J.; Zhang, X.; Cai, C.; and Li, K. 2017 b . Demoir \'e ing for screen-shot images with multi-channel layer decomposition. In 2017 IEEE Visual Communications and Image Processing (VCIP), 1--4. IEEE

    2017

  43. [51]

    Yu, J.; Li, X.; Koh, J.; Zhang, H.; Pang, R.; Qin, J.; Ku, A.; Xu, Y.; Baldridge, J.; and Wu, Y. 2021. Vector-quantized Image Modeling with Improved VQGAN. Cornell University - arXiv,Cornell University - arXiv

    2021

  44. [52]

    Yu, X.; Dai, P.; Li, W.; Ma, L.; Shen, J.; Li, J.; and Qi, X. 2022. Towards efficient and scale-robust ultra-high-definition image demoir \'e ing. In European Conference on Computer Vision, 646--662. Springer

    2022

  45. [53]

    Yuan, S.; Timofte, R.; Slabaugh, G.; Leonardis, A.; Zheng, B.; Ye, X.; Tian, X.; Chen, Y.; Cheng, X.; Fu, Z.; et al. 2019. Aim 2019 challenge on image demoireing: Methods and results. In 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), 3534--3545. IEEE

    2019

  46. [54]

    Yue, H.; Cheng, Y.; Mao, Y.; Cao, C.; and Yang, J. 2022. Recaptured screen image demoir \'e ing in raw domain. IEEE Transactions on Multimedia

    2022

  47. [55]

    Zhang, C.; Zhu, Y.; Yan, Q.; Sun, J.; and Zhang, Y. 2023. All-in-one multi-degradation image restoration network via hierarchical degradation representation. In Proceedings of the 31st ACM International Conference on Multimedia, 2285--2293

    2023

  48. [56]

    A.; Shechtman, E.; and Wang, O

    Zhang, R.; Isola, P.; Efros, A. A.; Shechtman, E.; and Wang, O. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, 586--595

    2018

  49. [57]

    Zheng, B.; Yuan, S.; Slabaugh, G.; and Leonardis, A. 2020. Image demoireing with learnable bandpass filters. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 3636--3645

    2020

  50. [58]

    Zheng, B.; Yuan, S.; Yan, C.; Tian, X.; Zhang, J.; Sun, Y.; Liu, L.; Leonardis, A.; and Slabaugh, G. 2021. Learning frequency domain priors for image demoireing. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(11): 7705--7717

    2021

  51. [59]

    Zhong, Y.; Zhou, Y.; Zhang, Y.; Chao, F.; and Ji, R. 2024. Learning Image Demoireing from Unpaired Real Data. arXiv preprint arXiv:2401.02719 (AAAI2024)

    2024 arXiv

  52. [60]

    Zhou, K.; Yang, Y.; Qiao, Y.; and Xiang, T. 2021. MixStyle Neural Networks for Domain Generalization and Adaptation. Cornell University - arXiv,Cornell University - arXiv

    2021

  53. [61]

    Zhu, Y.; Zhang, K.; Liang, J.; Cao, J.; Wen, B.; Timofte, R.; and Van Gool, L. 2023. Denoising Diffusion Models for Plug-and-Play Image Restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1219--1229

    2023

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.