Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Elucidating and Endowing the Diffusion Training Paradigm for General Image Restoration

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Transplanting the diffusion generative training objective into ordinary reconstruction-based restoration networks, guided by a per-degradation matching time step, improves generalization to unseen degradations and unifies ten restoration…

desk verdict A real framework with big empirical gains, but the matching time step tmat is a load-bearing premise whose transferability is not established. read the letter →

arxiv 2506.21722 v1 pith:OKMYQVUR submitted 2025-06-26 cs.CV cs.AI

Xin Lu , Xueyang Fu , Jie Xiao , Zihao Fan , Yurui Zhu , Zheng-Jun Zha This is my paper · ORCID
classification cs.CVcs.AI
keywords imagerestorationdiffusionmodelsgenerativepre-traininggeneralizationmulti-taskunifiedIRmatchingtimestepincrementaltrainingparameterregularization
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 sets out to prove that the diffusion training paradigm—the objective and schedule used to train generative diffusion models—can be grafted onto ordinary, reconstruction-based image restoration networks without changing their architecture or adding iterative sampling at inference. It claims that generative pre-training on clean images, followed by fine-tuning at a degradation-specific matching time step with regularization that preserves generative knowledge, sharply improves generalization to unseen degradation distributions in single-task restoration. For unified multi-task restoration, it claims that ordering incremental data introduction by the same matching time steps and adding time-prompted MoE adapters yields a single model that beats prior all-in-one methods across ten degradation types. If these claims hold, existing lightweight restoration backbones could inherit the robustness of diffusion models while keeping fast single-pass inference.

What carries the argument

The central organizing device is the degradation-specific matching time step $t_{\mathrm{mat}}$: for each degradation type, the pre-trained generative network runs the reverse diffusion process on a degraded image, and $t_{\mathrm{mat}}$ is the time step at which the successive-difference $(\mathbf{x}_t - \mathbf{x}_{t-1})$ best matches the degradation residual $(\mathbf{y} - \mathbf{x})$. This single scalar summarizes degradation complexity, selects the fine-tuning time step, orders the incremental multi-task training schedule, and supplies the time-based prompts to the MoE adapters. Around it the framework adds three regularizers: second-order parameter-importance regularization $L_{\mathrm{reg}}$ that slows forgetting of generative knowledge during fine-tuning, a gradient-orthogonality loss $L_{\mathrm{orthog}}$ that aligns generative and reconstruction gradient directions, and a layer-wise weight decay $w_{\mathrm{decay}}(t)=e^{-at}$ that confines generative fine-tuning to shallow, domain-invariant layers.

What would settle it

Compute the matching time step $t_{\mathrm{mat}}$ on one training split of a degradation (say RESIDE), then fine-tune the pre-trained network at $t_{\mathrm{mat}}$ and at neighboring time steps and evaluate on an out-of-distribution target (REVIDE); if $t_{\mathrm{mat}}$ does not coincide with the time step achieving the best target-domain PSNR, the framework's central premise fails. A cheaper check: re-estimate $t_{\mathrm{mat}}$ on a second training split or with a different training seed; if the estimated $t_{\mathrm{mat}}$ shifts by more than a few steps, it cannot be a stable degradation property.

Watch

Extended reading notes

Core claim

Diffusion generative pre-training on clean images gives an ordinary reconstruction-based restoration network $r_\theta(\mathbf{x}, t)$ access to the diffusion latent space, and each degradation type is assigned a matching time step $t_{\mathrm{mat}}$ via reverse sampling such that the reversed image aligns with the clean image. Fine-tuning the pre-trained network at that $t_{\mathrm{mat}}$, mixing about 10% generative data into the reconstruction fine-tuning, constraining updates with parameter-importance regularization and gradient orthogonality, and applying generative updates only to shallow layers produces restoration networks whose out-of-distribution generalization exceeds existing backbones in four single-task settings and whose unified model outperforms prior methods across ten degradations.

Load-bearing premise

The framework's gains ride on the assumption that the degradation-specific matching time step $t_{\mathrm{mat}}$, computed on training data, is a stable property of the degradation type that transfers to choosing fine-tuning time steps and ordering incremental training; if $t_{\mathrm{mat}}$ is instead a training-set artifact or needs tuning on the target generalization data, the advantage collapses.

Editorial extensions

If this is right

  • In single-task settings, the framework raises out-of-distribution PSNR by large margins, e.g., dehazing trained on RESIDE reaches 20.74 dB on REVIDE versus 17.26 dB for the best compared prior.
  • In unified multi-task settings, the framework reaches 28.13 dB average PSNR across ten degradations, outperforming the 27.51 dB of the best prior, with the best LPIPS among compared methods.
  • The framework is backbone-agnostic: applying it to NAFNet and X-Restormer lifts their unified-restoration averages to 28.47 and 28.32 dB respectively.
  • A mixing ratio of roughly 10% generative data during fine-tuning maximizes the benefit; larger generative ratios begin to degrade restoration fidelity.
  • Ordering incremental multi-task training by matching time steps gives better unified performance than directly mixing all task data.

Reading between the lines

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

  • If $t_{\mathrm{mat}}$ is indeed a stable property of each degradation type, it offers a cheap proxy for degradation complexity: one could predict relative transfer difficulty between degradations and schedule data collection or fine-tuning for new degradations without retraining, a program the paper does not itself pursue.
  • The shallow-layer/generative versus deep-layer/reconstruction split suggests a general recipe for injecting generative priors into discriminator-style networks, which may extend beyond restoration to super-resolution, JPEG artifact removal, or compressive sensing.
  • The paper reports $t_{\mathrm{mat}}$ increasing with global, complex degradations (low-light $t_{\mathrm{mat}}=38$, haze $t_{\mathrm{mat}}=47$, blur $t_{\mathrm{mat}}=50$), hinting at a monotone degradation-complexity axis; a testable extension is to verify this ordering on graded synthetic degradations.
  • The paper's stated limitation is extra training time; a natural follow-up is distilling the generative pre-training into fewer epochs or merging it into the fine-tuning stage, and the framework's own ablations suggest $L_{\mathrm{reg}}$ and $L_{\mathrm{orthog}}$ carry much of the benefit.
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

3 major / 5 minor

Summary. The paper proposes a training framework that transplants the diffusion generative training paradigm into general reconstruction-based image restoration (IR) networks. The method has three stages: Timestep Matched Generative Pre-training (TMGP) on clean images, Generalization Enhanced Fine-tuning (GEF) that mixes generative and reconstructive objectives with regularization losses, and Multi-task Unified Learning (MTUL) that uses a degradation-specific matching time step tmat to order incremental training and to condition MoE adapters. The authors report substantial single-task generalization improvements (e.g., dehazing generalization PSNR 20.74 vs. best prior 17.26 in Table 1) and state-of-the-art average multi-task performance on 10 degradations (Table 2, average PSNR 28.13 vs. RAM's 27.51), with ablations supporting each component and extensions to NAFNet and X-Restormer.

Significance. The paper's core idea—using the diffusion time-step structure as an organizing principle for IR training—is interesting and potentially useful for the community, especially because the framework is architecture-agnostic and the authors provide pseudo-code, extensive ablations, visualizations, and complexity comparisons. The strengths of the manuscript include the systematic exploratory analysis in Section 2, the breadth of experiments on 14 datasets, and the attempt to unify single-task generalization and multi-task unified IR under one training paradigm. However, the central premise that tmat is a stable, transferable property of a degradation type is not convincingly established, and the multi-task comparison protocol may not be equitable. If these issues are resolved, the framework could be a valuable training recipe for general IR networks.

major comments (3)
  1. [Section 3.1, Algorithm 2, Table 4, and Fig. 3(b,d)] The matching time step tmat is the load-bearing quantity of the framework, but its computation is under-specified and its transferability is not demonstrated. Algorithm 2 line 9 writes 'tmat ← t If min ∥(xt − xt−1) − (y − x)∥²' without specifying whether the minimization is over time steps, over samples in the training set, or over random realizations of the reverse-process noise z; since xt−1 depends on the sampled noise z in line 7, the returned tmat is stochastic, yet Table 4 reports single values with no variance or seed-level stability. More importantly, Fig. 3(b,d) sweeps the fine-tuning time step and evaluates on the target generalization sets (REVIDE, Rain100H, RealSnow, LOL-v2); this only shows that some time step chosen with access to the target data improves generalization, not that tmat computed from source training data alone would select that same step. Because tmat also determines the TIT ordering and the MoE prompts in Section 3.3, the entire framework's advantage rests on a premise that is currently unsupported. Please report tmat distributions over samples/seeds, verify that source-computed tmat matches the target-sweep optimum without target-set access, and compare TIT against alternative orderings (e.g., reverse, random, or complexity-based).
  2. [Section 4, Table 2, and Appendix G.1] The claim of superior multi-task unified IR performance requires that the Setting B baselines are trained on the same 10-task mixture, the same data splits, and a comparable training budget. The manuscript does not state whether the numbers for AirNet, Restormer, NAFNet, PromptIR, IR-SDE, DA-CLIP, ResShift, X-Restormer, and RAM are taken from their original papers or retrained under the authors' protocol. The data mixture described in Appendix G.1 is non-standard (e.g., a subset of Snow100K-L, CelebA-HQ for inpainting with RePaint masks, and a specific low-light set), so published numbers trained on different mixtures are not directly comparable. Please either retrain all baselines under the identical mixture and budget or clearly report the source of each baseline number and provide a matched-training comparison; otherwise the headline average PSNR gain of 0.62 dB over RAM could be an artifact of training-set composition rather than the proposed method.
  3. [Appendix F, Algorithm 6, and Eqs. (3)-(5)] The parameter importance regularization Lreg is not implementable as written. Equations (3)-(5) define Ωθk in terms of the post-fine-tuning parameters θ1, but the regularization loss is supposed to be used during fine-tuning before those parameters exist; the notation is also inconsistent (θ1_r vs. θ0_k vs. θ1_k). Algorithm 6 appears to compute δθ0_k = θ1_k − θ0_k only at the 'last training epoch' and then performs a single update θ1_k ← θ0_k + ε∇Ls, which does not correspond to normal gradient-descent training and would not produce the reported results. The ablation tables (Tables 6-10) show that Lreg contributes meaningfully to performance, so this component cannot remain a black box. Please provide a correct, implementable algorithm (e.g., an EWC-style importance computed at the end of pre-training and fixed during fine-tuning) and clarify how the Gauss-Newton approximation in Eq. (25) is used in practice.
minor comments (5)
  1. [Section 2.1, Fig. 4 and Fig. 5] Fig. 4(b) and Fig. 5 appear to show the same 'matching t guided incremental training' plots with nearly identical captions; please remove the duplicate or differentiate the two figures.
  2. [Throughout, equation numbering] The main text refers to 'Eq. 17', 'Eq. 18', and 'Eq. 19' for the gradient orthogonality terms, but these equations appear earlier in the main text as Eqs. (6)-(8); the appendix renumbers them as (17)-(19). Please use consistent equation numbering and correct all in-text references.
  3. [Algorithm 2, line 9] The condition 'If min ∥(xt − xt−1) − (y − x)∥²' is formally incomplete because the domain of minimization is not stated; specify whether the minimum is taken over t, over training samples, or over the sampled noise z.
  4. [Section G.2] The hyperparameter notation is inconsistent: the main text says 'λ = 0.2' while Appendix G.2 says 'α = 0.2'; please unify the notation.
  5. [Table 4 and Fig. 3(c)] Table 4 reports tmat values without error bars or statistics, and Fig. 3(c) presents claims about optimal network layers without numerical values or error bars; adding quantitative support would strengthen the conclusions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: tmat is an empirically selected training hyperparameter, not a prediction that reduces to its own inputs.

full rationale

The framework's load-bearing components—generative pre-training (Eq. 11/13), time-step-matched sampling (Algorithm 2), and GEF/MTUL training objectives (Eqs. 12, 16, 19, 22)—are defined independently of the reported evaluation metrics. tmat is computed by minimizing ||(xt−xt−1)−(y−x)|| on degradation pairs; though this is a data-derived value, the paper's claims that tmat transfers to fine-tuning and incremental ordering are empirical hypotheses tested against held-out generalization sets (e.g., RESIDE→REVIDE, Snow100K→RealSnow) and ablations (Tabs. 3, 6–10), not consequences of the definition. No equation in the paper reduces to its own output, no fitted parameter is renamed as a prediction of the same quantity, and self-citations (e.g., Refs. [20]–[23], [61]–[64]) are contextual and not load-bearing. Concerns about tmat stability or benchmark-specific selection are validity/risk issues rather than circularity, since the reported gains could in principle fail on new degradations without any self-referential contradiction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The framework's central claim rests on several fitted constants and empirical assumptions. The most consequential are the per-degradation matching time steps, which are selected by matching to clean images and then used both as prompts and as curriculum ordering, plus hyperparameters lambda, a, NE tuned on the benchmark. Diffusion background is standard, but the transfer assumptions (clean-image pre-training helps restoration; shallow layers are domain-invariant) are empirical and not independently grounded.

free parameters (5)
  • tmat per degradation type = 4, 8, 12, 15, 19, 22, 27, 38, 47, 50
    Selected by Algorithm 2 as the time step minimizing distance between reversed image and clean image on training data; used for fine-tuning time step and incremental training order.
  • lambda (Lreg balance) = 0.2
    Balance coefficient for parameter importance regularization; selected by ablation in Fig. 8.
  • a (weight decay rate) = 0.05
    Decay rate in wdecay(t) that controls layer-wise regularization; selected by ablation.
  • NE (number of MoE experts) = 10
    Number of experts in MoE adapters; selected by ablation in Fig. 8.
  • Mixing ratio of generative data = about 10%
    Ratio of generative training data during fine-tuning; chosen from Fig. 4 as a bottleneck point.
assumptions (4)
  • standard math Standard diffusion SDE/score matching/DDPM training background
    The paper relies on the standard DDPM and score-based diffusion formulations (Eqs. 1-2, Eq. 11).
  • domain assumption Generative pre-training on clean images transfers to degradation residuals
    The method pre-trains on clean images only and assumes this helps restoration fine-tuning; this is an empirical assumption validated only by experiments.
  • domain assumption Shallow layers are domain-invariant and deep layers are domain-specific
    The layer-wise regularization and weight decay design relies on feature visualizations in Fig. 9; this is an empirical claim.
  • ad hoc to paper tmat ordering corresponds to degradation complexity and curriculum helps
    The ordering of tasks by tmat values is interpreted as progressive complexity, but no independent evidence is given that this ordering is optimal beyond the reported experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Elucidating and Endowing the Diffusion Training Paradigm for General Image Restoration." pith.science (2026). https://pith.science/paper/OKMYQVUR

@misc{pith2026250621722,
  author       = {Pith},
  title        = {Pith review of: Elucidating and Endowing the Diffusion Training Paradigm for General Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OKMYQVUR}},
  note         = {Machine review of arXiv:2506.21722}
}
read the original abstract

While diffusion models demonstrate strong generative capabilities in image restoration (IR) tasks, their complex architectures and iterative processes limit their practical application compared to mainstream reconstruction-based general ordinary IR networks. Existing approaches primarily focus on optimizing network architecture and diffusion paths but overlook the integration of the diffusion training paradigm within general ordinary IR frameworks. To address these challenges, this paper elucidates key principles for adapting the diffusion training paradigm to general IR training through systematic analysis of time-step dependencies, network hierarchies, noise-level relationships, and multi-restoration task correlations, proposing a new IR framework supported by diffusion-based training. To enable IR networks to simultaneously restore images and model generative representations, we introduce a series of regularization strategies that align diffusion objectives with IR tasks, improving generalization in single-task scenarios. Furthermore, recognizing that diffusion-based generation exerts varying influences across different IR tasks, we develop an incremental training paradigm and task-specific adaptors, further enhancing performance in multi-task unified IR. Experiments demonstrate that our method significantly improves the generalization of IR networks in single-task IR and achieves superior performance in multi-task unified IR. Notably, the proposed framework can be seamlessly integrated into existing general IR architectures.

Figures

Figures reproduced from arXiv: 2506.21722 by the authors.

Figure 1
Figure 1. Visualizations of our method on image restoration (IR) tasks. When handling out-of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The top half of the figure illustrates the diffusion generation process, while the bottom half [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Generative pre-training and fine-tuning the pre-trained IR network with generative and reconstruction objectives (see [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (18 more)
Figure 6
Figure 6. Figure 6: Overall schema of our proposed method with diffusion’s generative training mechanism. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Ablation of loss and hyperparameters. for a simple yet effective approach to enhance the single-task generalization capability of general IR networks and improve the stability in addressing multi-task unified IR. Image restoration with diffusion. Diffusion models are p…
Figure 9
Figure 9. Figure 9: Comparison of shallow and deep features in generalization evaluations on synthetic and real datasets. The IR network extracts image information in shallow layer and outputs degraded residuals in deep layer. Our method reduces the damage to the background pixels of the …
Figure 10
Figure 10. Figure 10: T-SNE results of general IR Network X-Restormer’s feature (left) and feature of X￾Restormer [5] using our framework (right). We perform T-SNE analysis on multiple degradation types in Multi-task unified IR (Setting B). It can be found that general image restoration ne…
Figure 11
Figure 11. Figure 11: Comparison with existing diffusion-based approaches from a score-based perspective. (a) [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Example images with 10 image restoration tasks. For each task, the first row is the [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]
Figure 13
Figure 13. Figure 13: Visual performance of Single-task IR and generalization (Setting A) with 4 weather conditions. Our method can obtain cleaner images while preserving more background information when dealing with complex and multiple types of degradation. 23 [PITH_FULL_IMAGE:figures/f…
Figure 14
Figure 14. Figure 14: Image Inpainting visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: JPEG artifact deduction visualization comparisons of our method with previous approaches [PITH_FULL_IMAGE:figures/full_fig_p024_15.png]
Figure 16
Figure 16. Figure 16: Image deblurring visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]
Figure 17
Figure 17. Figure 17: Image deraining visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p025_17.png]
Figure 18
Figure 18. Figure 18: Image raindrop removal visualization comparisons of our method with previous approaches [PITH_FULL_IMAGE:figures/full_fig_p026_18.png]
Figure 19
Figure 19. Figure 19: Low-light image enhancement visualization comparisons of our method with previous [PITH_FULL_IMAGE:figures/full_fig_p026_19.png]
Figure 20
Figure 20. Figure 20: Visualization of our method for Multi-task unified IR (Setting B). Our method can obtain cleaner images while preserving more background information when dealing with complex and multiple types of degradation. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_20.png]
Figure 21
Figure 21. Figure 21: Visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p028_21.png]
Figure 22
Figure 22. Figure 22: Visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p029_22.png]
Figure 23
Figure 23. Figure 23: Visualization comparisons of our method with previous approaches for [PITH_FULL_IMAGE:figures/full_fig_p030_23.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AIM 2025 Challenge on High FPS Motion Deblurring: Methods and Results

    cs.CV 2025-09 conditional novelty 3.0 of 10

    The AIM 2025 challenge ranks 9 deblurring solutions on new high-FPS motion blur datasets, with VPEG placing first in both moderate and extreme tracks.

Reference graph

Works this paper leans on

79 extracted references · 71 canonical work pages · cited by 1 Pith paper

  1. [1]

    Agustsson and R

    E. Agustsson and R. Timofte. Ntire 2017 challenge on single image super-resolution: dataset and study. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 126–135, 2017

    work page 2017

  2. [2]

    B. D. O. Anderson. Reverse-time diffusion equation models. Stochastic Processes and Their Applications, 12(3):313–326, 1982

    work page 1982

  3. [3]

    L. Chen, X. Chu, X. Zhang, and J. Sun. Simple baselines for image restoration. InProceedings of the European Conference on Computer Vision (ECCV), page 17–33, Berlin, Heidelberg, 2022. Springer-Verlag. ISBN 978-3-031-20070-0. doi: 10.1007/978-3-031-20071-7_2. URL https://doi.org/10.1007/ 978-3-031-20071-7_2

    doi:10.1007/978-3-031-20071-7_2 2022

  4. [4]

    Chen, H.-Y

    W.-T. Chen, H.-Y . Fang, C.-L. Hsieh, C.-C. Tsai, I.-H. Chen, J.-J. Ding, and S.-Y . Kuo. All snow removed: Single image desnowing algorithm using hierarchical dual-tree complex wavelet representation and contradict channel loss. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4196–4205, October 2021

    work page 2021

  5. [5]

    X. Chen, Z. Li, Y . Pu, Y . Liu, J. Zhou, Y . Qiao, and C. Dong. A comparative study of image restoration networks for general backbone network design. In ECCV, 2024

    work page 2024

  6. [6]

    Robust Single Image Dehazing Based on Consistent and Contrast-Assisted Reconstruction

    D. Cheng, Y . Li, D. Zhang, N. Wang, X. Gao, and J. Sun. Robust single image dehazing based on consistent and contrast-assisted reconstruction. arXiv preprint arXiv:2203.15325, 2022

    work page Pith review arXiv 2022

  7. [7]

    H. Dong, J. shan Pan, L. Xiang, Z. Hu, X. Zhang, F. Wang, and M.-H. Yang. Multi-scale boosted dehazing network with dense feature fusion. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2154–2164, 2020. URL https://api.semanticscholar. org/CorpusID:216562731

    work page 2020

  8. [8]

    X. Fu, J. Huang, D. Zeng, Y . Huang, X. Ding, and J. Paisley. Removing rain from single images via a deep detail network. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1715–1723, 2017. doi: 10.1109/CVPR.2017.186

    work page doi:10.1109/cvpr.2017.186 2017

Show all 79 references
  1. [10]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models. In Proceedings of Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 6840–6851, 2020

    2020

  2. [11]

    Hyvärinen

    A. Hyvärinen. Estimation of non-normalized statistical models by score matching. Journal of Machine Learning Research, 6(4):695–709, 2005

    2005

  3. [12]

    Jiang, X

    Y . Jiang, X. Gong, D. Liu, Y . Cheng, C. Fang, X. Shen, J. Yang, P. Zhou, and Z. Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions on Image Processing, 30:2340–2349, 2021

    2021

  4. [13]

    Jiang, Z

    Y . Jiang, Z. Zhang, T. Xue, and J. Gu. Autodir: Automatic all-in-one image restoration with latent diffusion. In ECCV, 2024

    2024

  5. [14]

    Kirkpatrick, R

    J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Desjardins, A. A. Rusu, K. Milan, J. Quan, T. Ra- malho, A. Grabska-Barwinska, et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114(13):3521–3526, 2017

    2017

  6. [15]

    B. Li, W. Ren, D. Fu, D. Tao, D. Feng, W. Zeng, and Z. Wang. Benchmarking single-image dehazing and beyond. IEEE Transactions on Image Processing, 28(1):492–505, 2019

    2019

  7. [16]

    B. Li, X. Liu, P. Hu, Z. Wu, J. Lv, and X. Peng. All-in-one image restoration for unknown corruption. In CVPR, 2022

    2022

  8. [17]

    K. Liao, Z. Yue, Z. Wang, and C. C. Loy. Denoising as adaptation: Noise-space domain adaptation for image restoration. arXiv preprint arXiv:2406.18516, 2024

    2024 arXiv

  9. [18]

    X. Liu, Y . Ma, Z. Shi, and J. Chen. Griddehazenet: Attention-based multi-scale network for image dehazing. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7314–7323, 2019

    2019

  10. [19]

    Liu, D.-W

    Y .-F. Liu, D.-W. Jaw, S.-C. Huang, and J.-N. Hwang. Desnownet: Context-aware deep network for snow removal. IEEE Transactions on Image Processing, 27(6):3064–3073, 2018. 31

    2018

  11. [20]

    X. Lu, Y . Zhu, X. Wang, D. Li, J. Xiao, Y . Zhang, X. Fu, and Z.-J. Zha. Hirformer: Dynamic high resolution transformer for large-scale image shadow removal. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 6513–6523...

    2024

  12. [21]

    X. Lu, Y . Bao, J. Yang, A. Hu, J. Xiao, K. Wang, S. Xu, K. Liu, X. Fu, and Z.-J. Zha. Evenformer: Dynamic even transformer for real-world image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2025

    2025

  13. [22]

    X. Lu, J. Xiao, Y . Zhu, and X. Fu. Continuous adverse weather removal via degradation-aware distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2025

    2025

  14. [23]

    X. Lu, J. Yang, Y . Bao, A. Hu, Z. Fan, K. Wang, J. Xiao, X. Wang, X. Fu, and Z.-J. Zha. Advancing ambient lighting normalization via diffusion shadow generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2025

    2025

  15. [24]

    Lugmayr, M

    A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool. Repaint: inpainting using denoising diffusion probabilistic models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11461–11471, 2022

    2022

  16. [25]

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sjölund, and T. B. Schön. Image restoration with mean-reverting stochastic differential equations. International Conference on Machine Learning, 2023

    2023

  17. [26]

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sjölund, and T. B. Schön. Controlling vision-language models for universal image restoration. In ICLR, 2024

    2024

  18. [27]

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sjölund, and T. B. Schön. Controlling vision-language models for multi-task image restoration. In International Conference on Learning Representations, 2024

    2024

  19. [28]

    Z. Ma, Y . Zhang, G. Jia, L. Zhao, Y . Ma, M. Ma, G. Liu, K. Zhang, N. Ding, J. Li, and B. Zhou. Efficient diffusion models: A comprehensive survey from principles to practices. IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 1–20, 2025. doi: 10.1109/TPAM...

    2025

  20. [29]

    Martin, C

    D. Martin, C. Fowlkes, D. Tal, and J. Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In Proceedings of the 18th IEEE International Conference on Computer Vision (ICCV), volume 2...

    2001

  21. [30]

    McCloskey and N

    M. McCloskey and N. J. Cohen. Catastrophic interference in connectionist networks: The sequential learning problem. Psychology of Learning and Motivation, 24:109–165, 1989. URL https://api. semanticscholar.org/CorpusID:61019113

    1989

  22. [31]

    S. Nah, T. Hyun Kim, and K. Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 3883–3891, 2017

    2017

  23. [32]

    S. Nah, S. Son, J. Lee, and K. M. Lee. Clean images are hard to reblur: Exploiting the ill-posed inverse task for dynamic scene deblurring. In International Conference on Learning Representations, 2021. URL https://api.semanticscholar.org/CorpusID:247939726

    2021

  24. [33]

    Potlapalli, S

    V . Potlapalli, S. W. Zamir, S. H. Khan, and F. Shahbaz Khan. Promptir: Prompting for all-in-one image restoration. NeurIPS, 2024

    2024

  25. [34]

    R. Qian, R. T. Tan, W. Yang, J. Su, and J. Liu. Attentive generative adversarial network for raindrop removal from a single image. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2482–2491, 2017. URL https://api.semanticscholar.org/CorpusID:4539586

    2018

  26. [35]

    R. Qian, R. T. Tan, W. Yang, J. Su, and J. Liu. Attentive generative adversarial network for raindrop removal from a single image. In CVPR, 2018

    2018

  27. [36]

    Qin, R.-Q

    C.-J. Qin, R.-Q. Wu, Z. Liu, X. Lin, C.-L. Guo, H. H. Park, and C. Li. Restore anything with masks: Leveraging mask image modeling for blind all-in-one image restoration, 2024. URL https://arxiv. org/abs/2409.19403

    2024 arXiv

  28. [37]

    L. Qu, J. Tian, S. He, Y . Tang, and R. W. Lau. Deshadownet: A multi-context embedding deep network for shadow removal. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4067–4075, 2017. 32

    2017

  29. [38]

    Radford, K

    A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, et al. Improving language understanding by generative pre-training. San Francisco, CA, USA, 2018

    2018

  30. [39]

    C. Ren, X. He, C. Wang, and Z. Zhao. Adaptive consistency prior based deep network for image denoising. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 8592–8602,

    2021

  31. [40]

    D. Ren, W. Zuo, Q. Hu, P. Zhu, and D. Meng. Progressive image deraining networks: a better and simpler baseline. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3937–3946, 2019

    2019

  32. [41]

    W. Ren, S. Liu, H. Zhang, J. shan Pan, X. Cao, and M.-H. Yang. Single image dehazing via multi-scale convolutional neural networks. In European Conference on Computer Vision , 2016. URL https: //api.semanticscholar.org/CorpusID:17763780

    2016

  33. [42]

    Rissanen, M

    S. Rissanen, M. Heinonen, and A. Solin. Generative modelling with inverse heat dissipation. InProceedings of International Conference on Learning Representations (ICLR), 2022

    2022

  34. [43]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10684–10695, 2022

    2022

  35. [44]

    H. Sheikh. Live image quality assessment database release 2. http://live. ece. utexas. edu/research/quality, 2005

    2005

  36. [45]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning (ICML) , pages 2256–2265. PMLR, 2015

    2015

  37. [46]

    J. Song, C. Meng, and S. Ermon. Denoising diffusion implicit models. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021. URL https://openreview.net/forum?id=St1giarCHLP

    2021

  38. [47]

    J. Song, C. Meng, and S. Ermon. Denoising diffusion implicit models. In Proceedigns of International Conference on Learning Representations (ICLR), 2021

    2021

  39. [48]

    Song and S

    Y . Song and S. Ermon. Generative modeling by estimating gradients of the data distribution. InProceedings of Advances in Neural Information Processing Systems (NeurIPS), volume 32, 2019

    2019

  40. [49]

    Song and S

    Y . Song and S. Ermon. Improved techniques for training score-based generative models. InProceedings of Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 12438–12448, 2020

    2020

  41. [50]

    Y . Song, C. Durkan, I. Murray, and S. Ermon. Maximum likelihood training of score-based diffusion models. In Proceedings of Advances in Neural Information Processing Systems (NeurIPS), volume 34, pages 1415–1428, 2021

    2021

  42. [51]

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole. Score-based generative mod- eling through stochastic differential equations. In International Conference on Learning Representations (ICLR), 2021

    2021

  43. [52]

    Y . Song, Z. He, H. Qian, and X. Du. Vision transformers for single image dehazing.IEEE Transactions on Image Processing, 32:1927–1941, 2023

    1927

  44. [53]

    J. Tan, H. Yu, J. Huang, Z. Yang, and F. Zhao. Diffloss: unleashing diffusion model as constraint for training image restoration network. In Proceedings of the Asian Conference on Computer Vision, pages 1566–1584, 2024

    2024

  45. [54]

    Timofte, E

    R. Timofte, E. Agustsson, L. Van Gool, M.-H. Yang, and L. Zhang. NTIRE 2017 challenge on single image super-resolution: methods and results. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 114–125, 2017

    2017

  46. [55]

    Tsai, Y .-T

    F.-J. Tsai, Y .-T. Peng, C.-C. Tsai, Y .-Y . Lin, and C.-W. Lin. Banet: A blur-aware attention network for dynamic scene deblurring. IEEE Transactions on Image Processing , 31:6789–6799, 2021. URL https://api.semanticscholar.org/CorpusID:231639270

    2021

  47. [56]

    Z. Tu, H. Talebi, H. Zhang, F. Yang, P. Milanfar, A. Bovik, and Y . Li. MAXIM: Multi-axis MLP for image processing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5769–5780, 2022. 33

    2022

  48. [57]

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

    2004

  49. [58]

    C. Wei, W. Wang, W. Yang, and J. Liu. Deep retinex decomposition for low-light enhancement. arXiv preprint arXiv:1808.04560, 2018

    2018 arXiv

  50. [59]

    W. Wu, J. Weng, P. Zhang, X. Wang, W. Yang, and J. Jiang. Uretinex-net: Retinex-based deep unfolding network for low-light image enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5901–5910, 2022

    2022

  51. [60]

    B. Xia, Y . Zhang, S. Wang, Y . Wang, X. Wu, Y . Tian, W. Yang, and L. Van Gool. Diffir: Efficient diffusion model for image restoration. In CVPR, 2023

    2023

  52. [61]

    J. Xiao, M. Zhou, X. Fu, A. Liu, and Z.-J. Zha. Improving de-raining generalization via neural reorganiza- tion. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 4967–4976, 2021. doi: 10.1109/ICCV48922.2021.00494

    2021

  53. [62]

    J. Xiao, X. Fu, M. Zhou, H. Liu, and Z. Zha. Random shuffle transformer for image restoration. In International Conference on Machine Learning , 2023. URL https://api.semanticscholar. org/CorpusID:260957206

    2023

  54. [63]

    J. Xiao, R. Feng, H. Zhang, Z. Liu, Z. Yang, Y . Zhu, X. Fu, K. Zhu, Y . Liu, and Z. Zha. Dreamclean: Restoring clean image using deep diffusion prior. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 . OpenReview....

    2024

  55. [64]

    J. Xiao, X. Fu, Y . Zhu, D. Li, J. Huang, K. Zhu, and Z.-J. Zha. Homoformer: Homogenized transformer for image shadow removal. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 25617–25626, June 2024

    2024

  56. [65]

    H. Yang, L. Pan, Y . Yang, and W. Liang. Language-driven all-in-one adverse weather removal. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 24902–24912, June 2024

    2024

  57. [66]

    L. Yang, Z. Zhang, Y . Song, S. Hong, R. Xu, Y . Zhao, Y . Shao, W. Zhang, B. Cui, and M.-H. Yang. Diffusion models: a comprehensive survey of methods and applications. arXiv preprint arXiv:2209.00796, 2022

    2022

  58. [67]

    W. Yang, R. T. Tan, J. Feng, J. Liu, Z. Guo, and S. Yan. Deep joint rain detection and removal from a single image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), July 2017

    2017

  59. [68]

    W. Yang, R. T. Tan, J. Feng, J. Liu, Z. Guo, and S. Yan. Deep joint rain detection and removal from a single image. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pages 1357–1366, 2017

    2017

  60. [69]

    W. Yang, W. Wang, H. Huang, S. Wang, and J. Liu. Sparse gradient regularized deep retinex network for robust low-light image enhancement. IEEE Transactions on Image Processing, 30:2072–2086, 2021. URL https://api.semanticscholar.org/CorpusID:231641545

    2021

  61. [70]

    F. Yu, J. Gu, Z. Li, J. Hu, X. Kong, X. Wang, J. He, Y . Qiao, and C. Dong. Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild. In CVPR, 2024

    2024

  62. [72]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, M.-H. Yang, and L. Shao. Learning enriched features for real image restoration and enhancement. In Proceedings of the 16th European Conference on Computer Vision, pages 492–511. Springer, 2020

    2020

  63. [73]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, M.-H. Yang, and L. Shao. Multi-stage progres- sive image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14821–14831, 2021

    2021

  64. [74]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Yang. Restormer: efficient transformer for high-resolution image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5728–5739, 2022. 34

    2022

  65. [75]

    Zhang, W

    K. Zhang, W. Luo, Y . Zhong, L. Ma, B. Stenger, W. Liu, and H. Li. Deblurring by realistic blurring. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 2734–2743,

    2020

  66. [76]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 586–595, 2018

    2018

  67. [77]

    Zhang, H

    X. Zhang, H. Dong, J. Pan, C. Zhu, Y . Tai, C. Wang, J. Li, F. Huang, and F. Wang. Learning to restore hazy video: A new real-world dataset and a new method. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9235–9244, 2021. doi: 10.1109/CVPR...

    2021

  68. [78]

    Zheng, X.-M

    D. Zheng, X.-M. Wu, S. Yang, J. Zhang, J.-F. Hu, and W.-S. Zheng. Selective hourglass mapping for universal image restoration based on diffusion model. In CVPR, 2024

    2024

  69. [79]

    Y . Zhu, T. Wang, X. Fu, X. Yang, X. Guo, J. Dai, Y . Qiao, and X. hua Hu. Learning weather-general and weather-specific features for image restoration under multiple adverse weather conditions. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages ...

    2023

  70. [2020]

    doi: 10.1109/CVPR42600.2020.00281

    2020

  71. [2021]

    URL https://api.semanticscholar.org/CorpusID:235702920

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.