Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Complexity Experts are Task-Discriminative Learners for Any Image Restoration

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A cheap-first routing bias makes mixture-of-experts restoration networks task-specialized and state-of-the-art.

desk verdict Complexity-scaled experts with a parameter-count routing bias is a genuinely new idea, and the experiments largely support it, but the expert-scaling equation contradicts the text and needs fixing before the routing story can be fully trusted. read the letter →

arxiv 2411.18466 v2 pith:LCMMTXMW submitted 2024-11-27 cs.CV

Eduard Zamfir , Zongwei Wu , Nancy Mehta , Yuedong Tan , Danda Pani Paudel , Yulun Zhang , Radu Timofte This is my paper · ORCID
classification cs.CV
keywords all-in-oneimagerestorationmixture-of-expertscomplexityexpertstask-discriminativeroutingdehazingderainingdenoisingdynamicnetworks
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 introduces MoCE-IR, a single network for all-in-one image restoration, and argues that the key to making mixture-of-experts work there is to give the experts different computational complexity and receptive fields, then bias routing toward the cheaper ones. The authors' central claim is that this cheap-first bias, without any task labels, drives task-specific allocation: localized degradations such as rain land on small experts, while haze and noise, which need broader context, land on larger ones. They report state-of-the-art PSNR/SSIM on three, five, and composited degradation benchmarks at a lower average FLOP count than comparable models. A sympathetic reader should care because the recipe is nearly parameter-free, one auxiliary-loss bias computed from parameter counts, yet the experiments show it changes routing behavior qualitatively and improves restoration quality.

What carries the argument

The load-bearing mechanism is the mixture-of-complexity-experts (MoCE) layer, made of $n$ nested expert blocks whose embedding dimension shrinks as $r_i = C/2^i$ and whose window size grows, plus a shared transposed self-attention expert. A top-1 router picks one expert for the whole image under noisy softmax routing, and an auxiliary loss combines the coefficient of variation of expert importance and load, with importance multiplied by the complexity bias $b = [p_1/p_{\max}, \ldots, p_n/p_{\max}]$ based on each expert's learnable-parameter count. This bias is the only mechanism steering the router toward simpler experts, and the paper's ablations show it carries the task-discriminative behavior.

What would settle it

Train the same MoCE layer in the AIO-3 setting twice, once with the parameter-count bias and once with a bias computed from measured per-expert FLOPs or latency, and compare both routing histograms and average PSNR: if the task-to-expert mapping and the reported 32.57 dB average are unchanged, parameter count is an adequate complexity proxy, while if the mapping shifts or quality drops, the paper's causal story fails.

Watch

Extended reading notes

Core claim

The central discovery is that a mixture-of-experts layer built from non-uniform complexity experts, blocks with progressively smaller channel width and larger window partitions, spontaneously learns task-discriminative routing when the router's importance loss is scaled by a complexity bias $b = [p_1/p_{\max}, \ldots, p_n/p_{\max}]$. With image-level top-1 routing, the model assigns haze and denoising to high-capacity experts, rain to a lightweight local expert, and lets some experts serve several degradations at once. The paper shows this specialization is caused by the bias: replacing it with standard load balancing dissolves the specialization and lowers average PSNR from 32.57 dB to 32.30 dB on the three-degradation setting.

Load-bearing premise

The mechanism depends on parameter count per expert being a faithful proxy for computational complexity and for the receptive field a degradation needs, because only that count enters the routing bias.

Editorial extensions

If this is right

  • Inference can bypass high-complexity experts on easy inputs: the light model averages 36.93 GFLOPS and the heavy model 80.59 GFLOPS, with runtime around 22–23 ms at 224×224 on an RTX 4090.
  • The same network wins or ties the best previous results across all three benchmark groups: AIO-3 average PSNR 32.57 dB, AIO-5 average 30.58 dB, and CDD11 average 29.05 dB (light model), while using fewer parameters than most of its rivals.
  • Standard load balancing in place of the complexity bias drops AIO-3 average PSNR from 32.57 dB to 32.30 dB, which the paper presents as evidence that the bias, not the non-uniform experts alone, drives the gain.
  • Fixed post-training expert assignments show that complexity-biased routing produces larger performance differences between experts for a given task than load-balanced routing, so the router is genuinely task-discriminative.

Reading between the lines

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

  • A testable extension not explored here: replace the parameter-count bias with a FLOPs-based or latency-based bias and check whether the same task-to-expert mapping survives.
  • An implication the authors do not draw: because the router learns task identity without labels, the trained routing weights could serve as a lightweight degradation classifier.
  • A possible next step the paper mentions but does not test: patch-level routing for a single image containing multiple degradations, which image-level routing cannot handle.
  • If the cheap-first principle is general, inserting MoCE-style experts into other restoration backbones should reproduce the specialization, which would make the recipe portable.
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 / 6 minor

Summary. The paper proposes MoCE-IR, an all-in-one image restoration model in which each decoder block contains a mixture-of-experts layer with experts of different channel widths and window sizes, together with a shared expert. A complexity-aware auxiliary loss biases routing toward lower-complexity experts, with the stated goal of making task-specific allocation emerge automatically and allowing irrelevant experts to be bypassed at inference. The authors validate the method on three-degradation, five-degradation, and composited-degradation benchmarks, report gains over AirNet, PromptIR, InstructIR, OneRestore, and other baselines at lower parameter counts, and provide ablations on expert scaling, routing bias, routing visualizations, and fixed-expert analysis.

Significance. The empirical scope is a strength: evaluations cover AIO-3, AIO-5, and CDD11 composite settings at two model scales, with ablations isolating the complexity bias (Tab. 5b), routing visualizations (Fig. 5), and a fixed-expert test (Tab. 6). The code and models are promised publicly, and the efficiency measurements in Tab. 4 directly address the practical motivation of bypassing experts. If the central mechanism is as described, the paper would make a useful contribution by showing that a simple parameter-count bias in the load-balancing loss induces task-specific routing without explicit degradation labels. However, the current manuscript contains an internal contradiction in the expert complexity schedule that affects the interpretation of the claimed bias, and the quantitative claims would benefit from repeated-run statistics.

major comments (3)
  1. [Sec. 3.1, Eq. (2); Sec. 3.2, Eq. (6)] The expert-complexity ordering is internally inconsistent. The text says the most lightweight expert has the smallest embedding dimension r1 and window size w1, but Eq. (2) defines r = C/2^i, so for i=1,...,n expert 1 has the largest embedding (r1 = C/2) and expert n the smallest (rn = C/2^n). Because the projection layers in Eq. (2) are linear in r, parameter count p_i in Eq. (6) decreases with i, which reverses the labels used in the routing discussion and in Fig. 5, where the y-axis is labeled 'increasing expert complexity'. Moreover, parameter count and receptive field are traded off (r decreases while w increases), so the bias b_i = p_i/p_max is not an unambiguous complexity proxy. The authors must state the implemented schedule (e.g., r_i = C/2^{n-i+1}), verify it against the released code, and either justify p_i as the complexity measure or report per-expert FLOPs.
  2. [Sec. 4, Tables 1-4] All quantitative comparisons are reported as single PSNR/SSIM values with no repeated runs or confidence intervals. Several of the claimed gains are small (e.g., 0.02-0.04 dB over UniProcessor in Tab. 1 and 0.03 dB over InstructIR in Tab. 2), so without variance estimates it is not possible to judge whether the differences are significant. Please add at least three seeds for the main tables or report the training variance, and clarify how the mean and standard deviation of FLOPS/runtime in Tab. 4 were computed (which tasks, how many runs, and whether router overhead is included).
  3. [Sec. 4.2, Table 6 and Fig. 5] The fixed-expert analysis that supports the task-discriminative claim is incomplete. Table 6 reports only rain and noise, omitting haze, and the learned routing for noise (E4, 33.92 dB) is actually slightly worse than the manual choice E4 (34.00 dB), which weakens the claim that learned routing finds the best expert. The rows for the load-balancing baseline are marked 'Not Applicable' without explanation. Please extend the analysis to all degradations and report the full routing matrix per layer/task, or temper the conclusion that the method 'automatically' assigns every task to its most suitable expert.
minor comments (6)
  1. [Sec. 3.1] Equation (2) and the implementation-details paragraph both write 'i ∈ {i, ..., n}'; this should be 'i ∈ {1, ..., n}'.
  2. [Sec. 3.2] There is a typo in 'Withing(x), we select...'; it should be 'Within g(x), ...'.
  3. [Tables 5b and 6] The load-balancing baseline is cited as [45] in Tables 5b and 6, but reference [45] is an augmented-reality article; the load-balancing loss in the text is attributed to Riquelme et al. [43], so the table citations should be corrected.
  4. [Sec. 4.2, Tab. 5a] The exact definitions of 'nested' and 'exponential' expert scaling, and of the '2(2+i)' window progression, are not given; please provide the precise formulas.
  5. [Fig. 5] The caption should define what is averaged in the routing heatmaps (images, tokens, or layers) and specify the color scale units.
  6. [Introduction] The introduction contains a typo: 'Noteable works' should be 'Notable works'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the complexity-bias routing is an explicit architectural prior, and the claimed task-discriminative allocation is an empirically evaluated learned outcome, not a restatement of the loss.

full rationale

The paper's central mechanism is a fixed complexity bias b = [p_i/p_max] (Eq. 6) that reweights the standard load-balancing importance term (Eq. 5), and task-specific expert selection is learned under this prior. The bias is not fitted to the reported PSNR/SSIM numbers, and no 'prediction' in Tables 1-3 is constructed from the same data that defines the method's free parameters; the routing patterns in Fig. 5 are empirical readouts, not identities. The ablations in Tab. 5b compare against standard load balancing [43] and other routers, which is the correct control and shows the bias changes outcomes rather than merely renaming them. Self-citations to the authors' prior MoE restoration work ([62], [63]) appear only as related-work context (e.g., 'recent all-in-one models [29, 57, 62] leverage various priors for expert routing') and are not load-bearing for the MoCE contribution. The paper does contain an internal-consistency issue in Sec. 3.1: the printed formula r = C/2^i makes expert 1 the largest in channel dimension while the text calls r1 and w1 the most lightweight; this is a correctness/ambiguity concern about the complexity ordering, not a circular derivation, because the loss and routing would behave according to whichever schedule is actually implemented. The Limitations section acknowledges scalability constraints of image-level routing, which is an honest scope statement. Overall, the central claim is independently testable against external restoration benchmarks and against the load-balancing baseline, so there is no circularity.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

No new physical entities are postulated. The 'complexity experts' are architectural components, not entities with independent falsifiable predictions.

free parameters (7)
  • Number of experts n = 4
    Set to 4 by hand; performance likely depends on this.
  • Expert channel scaling = nested r = C/2^i (ambiguous, see text)
    Selected via ablation (Table 5a); the formula conflicts with the text description.
  • Bias normalization = pMax
    Chosen via ablation; normalizing by pMin hurts performance.
  • Top-k in routing = 1
    Hard-coded; standard for sparse MoE.
  • Noise variance for routing = 1/n^2
    Chosen by hand.
  • Auxiliary loss coefficient = 1/2
    Chosen by hand in Equation 7.
  • Window sizes w_i = not specified numerically
    The paper says window size increases but does not report the exact values, making exact reproduction impossible.
assumptions (5)
  • domain assumption FFT-based approximation of self-attention preserves the intended receptive-field differences across experts.
    Sec. 3.1 Eq. 2, based on FFTformer [21]; if the approximation collapses the window-size distinctions, complexity experts may not differ as intended.
  • domain assumption Transposed self-attention in the channel dimension serves as an adequate task-agnostic shared expert.
    Sec. 3.1 Eq. 3, taken from Restormer [65].
  • ad hoc to paper Parameter count is a valid proxy for expert complexity in the routing bias.
    Eq. 6 uses p_i/p_max; this ignores window size and actual FLOPs, so the bias may not match the intended complexity ordering.
  • domain assumption Image-level routing (one expert per image per layer) is sufficient for all-in-one restoration.
    Sec. 3.2; acknowledged in the limitations, but it limits fine-grained spatial routing.
  • domain assumption The tested degradation types and datasets are representative of 'any image restoration' as claimed in the title.
    Experiments cover 3, 5, and 11 settings, but not super-resolution, inpainting, or compression artifacts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Complexity Experts are Task-Discriminative Learners for Any Image Restoration." pith.science (2026). https://pith.science/paper/LCMMTXMW

@misc{pith2026241118466,
  author       = {Pith},
  title        = {Pith review of: Complexity Experts are Task-Discriminative Learners for Any Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LCMMTXMW}},
  note         = {Machine review of arXiv:2411.18466}
}
read the original abstract

Recent advancements in all-in-one image restoration models have revolutionized the ability to address diverse degradations through a unified framework. However, parameters tied to specific tasks often remain inactive for other tasks, making mixture-of-experts (MoE) architectures a natural extension. Despite this, MoEs often show inconsistent behavior, with some experts unexpectedly generalizing across tasks while others struggle within their intended scope. This hinders leveraging MoEs' computational benefits by bypassing irrelevant experts during inference. We attribute this undesired behavior to the uniform and rigid architecture of traditional MoEs. To address this, we introduce ``complexity experts" -- flexible expert blocks with varying computational complexity and receptive fields. A key challenge is assigning tasks to each expert, as degradation complexity is unknown in advance. Thus, we execute tasks with a simple bias toward lower complexity. To our surprise, this preference effectively drives task-specific allocation, assigning tasks to experts with the appropriate complexity. Extensive experiments validate our approach, demonstrating the ability to bypass irrelevant experts during inference while maintaining superior performance. The proposed MoCE-IR model outperforms state-of-the-art methods, affirming its efficiency and practical applicability. The source code and models are publicly available at \href{https://eduardzamfir.github.io/moceir/}{\texttt{eduardzamfir.github.io/MoCE-IR/}}

Figures

Figures reproduced from arXiv: 2411.18466 by the authors.

Figure 1
Figure 1. Motivation. (a) Dense all-in-one restoration meth￾ods [24, 57] often inefficiently allocate parameters when handling multiple degradation types. (b) While recent Mixture-of-Experts (MoE) approaches [61, 62] address this through sparse computa￾tion, their rigid routing mechanisms uniformly distribute inputs across experts without considering the natural relationships be￾tween degradations. (c) To overcome these limit… view at source ↗
Figure 2
Figure 2. Proposed MoCE-IR framework. Despite recent advances in MoE-based image restoration [57, 62, 63], inconsistent expert behavior—where some experts over-generalize while others underperform—limits their computational efficiency. We address this through complexity experts: flexible blocks with varying computational capacity and receptive fields. Our MoCE-IR employs an asymmetric encoder-decoder architecture where each d… view at source ↗
Figure 3
Figure 3. Visual results. We compare MoCE-IR-S to AirNet [24], and PromptIR [36] in the all-in-one setting with three degradations. MoCE-IR-S effectively removes haze and rain streaks while preserving image sharpness, achieving high-quality restoration. An error heatmap is provided, with color transitioning from black to white to indicate increasing pixel-wise error. 4.1. Comparison to State-of-the-Art Methods All-in-One: Thr… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Complexity-efficiency tradeoff. Visualization of PSNR and parameter counts of proposed method compared to prior work. Proposed MoCE-IR surpasses prior methods, achieving SoTA re￾sults in all-in-one image restoration with enhanced efficiency [PITH_FULL_IMAGE:figures/fu…
Figure 5
Figure 5. Figure 5: Routing visualization for the AIO-3 setting. (a) While load balancing [43] ensures uniform expert utilization, it neglects shared task dependencies and task-specific characteristics, limiting restoration quality. (b)-(d) Complexity-aware routing fosters task discrimina…

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. M2Restore: Mixture-of-Experts-based Mamba-CNN Fusion Framework for All-in-One Image Restoration

    cs.CV 2025-06 conditional novelty 5.0 of 10

    M2Restore is a CLIP-guided Mixture-of-Experts Mamba-CNN model that reports state-of-the-art results on the All-weather all-in-one image restoration benchmark.

Reference graph

Works this paper leans on

72 extracted references · 59 canonical work pages · cited by 1 Pith paper

  1. [1]

    Multimodal prompt perceiver: Empower adap- tiveness generalizability and fidelity for all-in-one image restoration

    Yuang Ai, Huaibo Huang, Xiaoqiang Zhou, Jiexiang Wang, and Ran He. Multimodal prompt perceiver: Empower adap- tiveness generalizability and fidelity for all-in-one image restoration. In CVPR, 2024. 2

    2024

  2. [2]

    Contour detection and hierarchical image segmen- tation

    Pablo Arbelaez, Michael Maire, Charless Fowlkes, and Jiten- dra Malik. Contour detection and hierarchical image segmen- tation. TPAMI, 33(5):898–916, 2010. 5

    work page 2010

  3. [3]

    Es- timating or propagating gradients through stochastic neurons for conditional computation

    Yoshua Bengio, Nicholas L´eonard, and Aaron Courville. Es- timating or propagating gradients through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432,

    arXiv

  4. [4]

    Non-local image dehazing

    Dana Berman and Shai Avidan. Non-local image dehazing. In CVPR, 2016. 2

    work page 2016

  5. [5]

    Dehazenet: An end-to-end system for single image haze removal

    Bolun Cai, Xiangmin Xu, Kui Jia, Chunmei Qing, and Dacheng Tao. Dehazenet: An end-to-end system for single image haze removal. TIP, 25(11):5187–5198, 2016. 1

    work page 2016

  6. [6]

    Simple baselines for image restoration

    Liangyu Chen, Xiaojie Chu, Xiangyu Zhang, and Jian Sun. Simple baselines for image restoration. In ECCV, 2022. 1, 2, 5

    work page 2022

  7. [7]

    Always Clear Days: Degradation Type and Severity Aware All-In-One Adverse Weather Removal

    Yu-Wei Chen and Soo-Chang Pei. Always clear days: Degra- dation type and severity aware all-in-one adverse weather removal. arXiv preprint arXiv:2310.18293, 2023. 1, 2

    work page Pith review arXiv 2023

  8. [8]

    Cross aggregation transformer for image restoration

    Zheng Chen, Yulun Zhang, Jinjin Gu, Linghe Kong, Xin Yuan, et al. Cross aggregation transformer for image restoration. NeurIPS, 2022. 1, 2

    work page 2022

Show all 72 references
  1. [9]

    In- structir: High-quality image restoration following human instructions

    Marcos V Conde, Gregor Geigle, and Radu Timofte. In- structir: High-quality image restoration following human instructions. In ECCV, 2024. 1, 2, 3, 5, 6

    2024

  2. [10]

    An application performance optimization model of mobile augmented reality based on hd restoration

    Xin Dang, Hai Wang, Jie Ren, and Le Chen. An application performance optimization model of mobile augmented reality based on hd restoration. In CBD, pages 201–206. IEEE, 2020. 1

    2020

  3. [11]

    Fd-gan: Generative adversarial networks with fusion- discriminator for single image dehazing

    Yu Dong, Yihao Liu, He Zhang, Shifeng Chen, and Yu Qiao. Fd-gan: Generative adversarial networks with fusion- discriminator for single image dehazing. In AAAI, 2020. 5

    2020

  4. [12]

    Uniprocessor: A text-induced unified low- level image processor

    Huiyu Duan, Xiongkuo Min, Sijing Wu, Wei Shen, and Guangtao Zhai. Uniprocessor: A text-induced unified low- level image processor. In ECCV, 2024. 1, 3, 5, 6

    2024

  5. [13]

    A general decoupled learning framework for parameterized image operators

    Qingnan Fan, Dongdong Chen, Lu Yuan, Gang Hua, Neng- hai Yu, and Baoquan Chen. A general decoupled learning framework for parameterized image operators. TPAMI, 43(1): 33–47, 2019. 1, 5

    2019

  6. [14]

    Dy- namic scene deblurring with parameter selective sharing and nested skip connections

    Hongyun Gao, Xin Tao, Xiaoyong Shen, and Jiaya Jia. Dy- namic scene deblurring with parameter selective sharing and nested skip connections. In CVPR, 2019. 5

    2019

  7. [15]

    Virtual restoration of deteriorated reli- gious heritage objects using augmented reality technologies

    Florin Girbacia, Silviu Butnariu, Alex Petre Orman, and Cris- tian Cezar Postelnicu. Virtual restoration of deteriorated reli- gious heritage objects using augmented reality technologies. European Journal of Science and Theology , 9(2):223–231,

  8. [16]

    Mambair: A simple baseline for image restoration with state-space model

    Hang Guo, Jinmin Li, Tao Dai, Zhihao Ouyang, Xudong Ren, and Shu-Tao Xia. Mambair: A simple baseline for image restoration with state-space model. In ECCV, 2024. 1, 5

    2024

  9. [17]

    Onerestore: A universal restoration framework for com- posite degradation

    Yu Guo, Yuan Gao, Yuxu Lu, Ryan Wen Liu, and Shengfeng He. Onerestore: A universal restoration framework for com- posite degradation. In ECCV, 2024. 5, 6, 7

    2024

  10. [18]

    Multi-scale progressive fusion network for single image deraining

    Kui Jiang, Zhongyuan Wang, Peng Yi, Chen Chen, Baojin Huang, Yimin Luo, Jiayi Ma, and Junjun Jiang. Multi-scale progressive fusion network for single image deraining. In CVPR, 2020. 2

    2020

  11. [19]

    Autodir: Automatic all-in-one image restoration with latent diffusion

    Yitong Jiang, Zhaoyang Zhang, Tianfan Xue, and Jinwei Gu. Autodir: Automatic all-in-one image restoration with latent diffusion. arXiv preprint arXiv:2310.10123, 2023. 2

    2023 arXiv

  12. [20]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv, 2017. 5

    2017

  13. [21]

    Efficient frequency domain-based trans- formers for high-quality image deblurring

    Lingshun Kong, Jiangxin Dong, Jianjun Ge, Mingqiang Li, and Jinshan Pan. Efficient frequency domain-based trans- formers for high-quality image deblurring. In CVPR, 2023. 2, 4

    2023

  14. [22]

    Noise2noise: Learning image restoration without clean data

    Jaakko Lehtinen, Jacob Munkberg, Jon Hasselgren, Samuli Laine, Tero Karras, Miika Aittala, and Timo Aila. Noise2noise: Learning image restoration without clean data. arXiv preprint arXiv:1803.04189, 2018. 1, 2

    2018 arXiv

  15. [23]

    Benchmarking single- image dehazing and beyond

    Boyi Li, Wenqi Ren, Dengpan Fu, Dacheng Tao, Dan Feng, Wenjun Zeng, and Zhangyang Wang. Benchmarking single- image dehazing and beyond. TIP, 28(1):492–505, 2018. 5

    2018

  16. [24]

    All-In-One Image Restoration for Unknown Corruption

    Boyun Li, Xiao Liu, Peng Hu, Zhongqin Wu, Jiancheng Lv, and Xi Peng. All-In-One Image Restoration for Unknown Corruption. In CVPR, 2022. 1, 2, 5, 6, 7

    2022

  17. [25]

    Prompt-in-prompt learning for universal image restoration

    Zilong Li, Yiming Lei, Chenglong Ma, Junping Zhang, and Hongming Shan. Prompt-in-prompt learning for universal image restoration. arXiv preprint arXiv:2312.05038, 2023. 1, 2

    2023 arXiv

  18. [26]

    Swinir: Image restoration using swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. In ICCV, 2021. 2, 5

    2021

  19. [27]

    Tape: Task-agnostic prior embedding for image restoration

    Lin Liu, Lingxi Xie, Xiaopeng Zhang, Shanxin Yuan, Xi- angyu Chen, Wengang Zhou, Houqiang Li, and Qi Tian. Tape: Task-agnostic prior embedding for image restoration. In ECCV, 2022. 1, 5

    2022

  20. [28]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In ICCV, 2021. 1, 2

    2021

  21. [29]

    Wm-moe: Weather-aware multi-scale mixture-of-experts for blind adverse weather removal

    Yulin Luo, Rui Zhao, Xiaobao Wei, Jinwei Chen, Yijie Lu, Shenghao Xie, Tianyu Wang, Ruiqin Xiong, Ming Lu, and Shanghang Zhang. Wm-moe: Weather-aware multi-scale mixture-of-experts for blind adverse weather removal. arXiv preprint arXiv:2303.13739, 2023. 2, 3, 7, 8

    2023 arXiv

  22. [30]

    Controlling vision-language models for multi-task image restoration

    Ziwei Luo, Fredrik K Gustafsson, Zheng Zhao, Jens Sj¨olund, and Thomas B Sch¨on. Controlling vision-language models for multi-task image restoration. In ICLR, 2024. 2, 3, 5, 6

    2024

  23. [31]

    Waterloo exploration database: New challenges for image quality as- sessment models

    Kede Ma, Zhengfang Duanmu, Qingbo Wu, Zhou Wang, Hongwei Yong, Hongliang Li, and Lei Zhang. Waterloo exploration database: New challenges for image quality as- sessment models. TIP, 26(2):1004–1016, 2016. 5

    2016

  24. [32]

    A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics

    David Martin, Charless Fowlkes, Doron Tal, and Jitendra Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In ICCV, 2001. 5

    2001

  25. [33]

    Deep generalized unfolding networks for image restoration

    Chong Mou, Qian Wang, and Jian Zhang. Deep generalized unfolding networks for image restoration. In CVPR, 2022. 5 9

    2022

  26. [34]

    Deep multi-scale convolutional neural network for dynamic scene deblurring

    Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. In CVPR, 2017. 5

    2017

  27. [35]

    Blind image deblurring using dark channel prior

    Jinshan Pan, Deqing Sun, Hans Pfister, and Ming-Hsuan Yang. Blind image deblurring using dark channel prior. In CVPR,

  28. [36]

    Promptir: Prompting for all-in-one image restoration

    Vaishnav Potlapalli, Syed Waqas Zamir, Salman Khan, and Fahad Khan. Promptir: Prompting for all-in-one image restoration. In NeurIPS, 2023. 1, 2, 3, 5, 6, 7

    2023

  29. [37]

    From sparse to soft mixtures of experts

    Joan Puigcerver, Carlos Riquelme Ruiz, Basil Mustafa, and Neil Houlsby. From sparse to soft mixtures of experts. In ICLR, 2024. 2, 3, 4, 8

    2024

  30. [38]

    En- hanced pix2pix dehazing network

    Yanyun Qu, Yizi Chen, Jingying Huang, and Yuan Xie. En- hanced pix2pix dehazing network. In CVPR, 2019. 1

    2019

  31. [39]

    Mixture-of- depths: Dynamically allocating compute in transformer-based language models, 2024

    David Raposo, Sam Ritter, Blake Richards, Timothy Lillicrap, Peter Conway Humphreys, and Adam Santoro. Mixture-of- depths: Dynamically allocating compute in transformer-based language models, 2024. 2, 8

    2024

  32. [40]

    Progressive image deraining networks: A better and simpler baseline

    Dongwei Ren, Wangmeng Zuo, Qinghua Hu, Pengfei Zhu, and Deyu Meng. Progressive image deraining networks: A better and simpler baseline. In CVPR, 2019. 2

    2019

  33. [41]

    Gated fusion network for single image dehazing

    Wenqi Ren, Lin Ma, Jiawei Zhang, Jinshan Pan, Xiaochun Cao, Wei Liu, and Ming-Hsuan Yang. Gated fusion network for single image dehazing. In CVPR, 2018. 2

    2018

  34. [42]

    Single image dehazing via multi-scale convolutional neural networks with holistic edges

    Wenqi Ren, Jinshan Pan, Hua Zhang, Xiaochun Cao, and Ming-Hsuan Yang. Single image dehazing via multi-scale convolutional neural networks with holistic edges. In ICCV,

  35. [43]

    Scaling vision with sparse mixture of experts

    Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, Andr´e Susano Pinto, Daniel Keysers, and Neil Houlsby. Scaling vision with sparse mixture of experts. NeurIPS, 2021. 2, 3, 4, 7, 8

    2021

  36. [44]

    U-Net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-Net: Convolutional networks for biomedical image segmentation. In MICCAI, pages 234–241. Springer, 2015. 3

    2015

  37. [45]

    Augmented reality for restoration/reconstruction of artefacts with artistic or histori- cal value

    Giovanni Saggio, Davide Borra, et al. Augmented reality for restoration/reconstruction of artefacts with artistic or histori- cal value. In Augmented reality: some emerging application areas, pages 59–86. InTech Publications, 2011. 1

    2011

  38. [46]

    Outra- geously large neural networks: The sparsely-gated mixture- of-experts layer, 2017

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outra- geously large neural networks: The sparsely-gated mixture- of-experts layer, 2017. 3, 4

    2017

  39. [47]

    Mem- net: A persistent memory network for image restoration

    Ying Tai, Jian Yang, Xiaoming Liu, and Chunyan Xu. Mem- net: A persistent memory network for image restoration. In ICCV, 2017. 1, 2

    2017

  40. [48]

    Image denois- ing using deep cnn with batch renormalization

    Chunwei Tian, Yong Xu, and Wangmeng Zuo. Image denois- ing using deep cnn with batch renormalization. NN, 2020. 5

    2020

  41. [49]

    Transweather: Transformer-based restoration of images degraded by adverse weather conditions

    Jeya Maria Jose Valanarasu, Rajeev Yasarla, and Vishal M Patel. Transweather: Transformer-based restoration of images degraded by adverse weather conditions. In CVPR, 2022. 1, 2, 5

    2022

  42. [50]

    Promptre- storer: A prompting image restoration method with degrada- tion perception

    Cong Wang, Jinshan Pan, Wei Wang, Jiangxin Dong, Mengzhu Wang, Yakun Ju, and Junyang Chen. Promptre- storer: A prompting image restoration method with degrada- tion perception. In NeurIPS, 2023. 1, 2

    2023

  43. [51]

    Gridformer: Residual dense transformer with grid structure for image restoration in adverse weather conditions

    Tao Wang, Kaihao Zhang, Ziqian Shao, Wenhan Luo, Bjorn Stenger, Tong Lu, Tae-Kyun Kim, Wei Liu, and Hongdong Li. Gridformer: Residual dense transformer with grid structure for image restoration in adverse weather conditions. ICCV,

  44. [52]

    Uformer: A general u-shaped transformer for image restoration

    Zhendong Wang, Xiaodong Cun, Jianmin Bao, Wengang Zhou, Jianzhuang Liu, and Houqiang Li. Uformer: A general u-shaped transformer for image restoration. In CVPR, 2022. 1, 2

    2022

  45. [53]

    Deep retinex decomposition for low-light enhancement.arXiv preprint arXiv:1808.04560, 2018

    Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. Deep retinex decomposition for low-light enhancement.arXiv preprint arXiv:1808.04560, 2018. 5

    2018 arXiv

  46. [54]

    Har- mony in diversity: Improving all-in-one image restoration via multi-task collaboration

    Gang Wu, Junjun Jiang, Kui Jiang, and Xianming Liu. Har- mony in diversity: Improving all-in-one image restoration via multi-task collaboration. In ACMMM, 2024. 5

    2024

  47. [55]

    Contrastive learning for compact single image dehazing

    Haiyan Wu, Yanyun Qu, Shaohui Lin, Jian Zhou, Ruizhi Qiao, Zhizhong Zhang, Yuan Xie, and Lizhuang Ma. Contrastive learning for compact single image dehazing. In CVPR, 2021. 2

    2021

  48. [56]

    Learning texture transformer network for image super-resolution

    Fuzhi Yang, Huan Yang, Jianlong Fu, Hongtao Lu, and Bain- ing Guo. Learning texture transformer network for image super-resolution. In CVPR, 2020. 5

    2020

  49. [57]

    Language- driven all-in-one adverse weather removal

    Hao Yang, Liyuan Pan, Yan Yang, and Wei Liang. Language- driven all-in-one adverse weather removal. In CVPR, 2024. 1, 2, 3

    2024

  50. [58]

    All-in-one medical image restoration via task-adaptive routing

    Zhiwen Yang, Haowei Chen, Ziniu Qian, Yang Yi, Hui Zhang, Dan Zhao, Bingzheng Wei, and Yan Xu. All-in-one medical image restoration via task-adaptive routing. In MICCAI, 2024. 7, 8

    2024

  51. [59]

    Uncertainty guided multi- scale residual learning-using a cycle spinning cnn for single image de-raining

    Rajeev Yasarla and Vishal M Patel. Uncertainty guided multi- scale residual learning-using a cycle spinning cnn for single image de-raining. In CVPR, 2019. 1

    2019

  52. [60]

    Path-restore: Learning network path se- lection for image restoration

    Ke Yu, Xintao Wang, Chao Dong, Xiaoou Tang, and Chen Change Loy. Path-restore: Learning network path se- lection for image restoration. TPAMI, 2022. 3

    2022

  53. [61]

    Multi-expert adaptive selection: Task-balancing for all-in-one image restoration

    Xiaoyan Yu, Shen Zhou, Huafeng Li, and Liehuang Zhu. Multi-expert adaptive selection: Task-balancing for all-in-one image restoration. arXiv preprint arXiv:2407.19139, 2024. 1

    2024 arXiv

  54. [62]

    Efficient degradation-aware any image restoration

    Eduard Zamfir, Zongwei Wu, Nancy Mehta, Danda Pani Paudel, Yulun Zhang, and Radu Timofte. Efficient degradation-aware any image restoration. arXiv preprint arXiv:2405.15475, 2024. 1, 2, 3

    2024 arXiv

  55. [63]

    See more details: Efficient image super- resolution by experts mining

    Eduard Zamfir, Zongwei Wu, Nancy Mehta, Yulun Zhang, and Radu Timofte. See more details: Efficient image super- resolution by experts mining. In ICML. PMLR, 2024. 3

    2024

  56. [64]

    Multi-stage progressive image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shahbaz Khan, Ming-Hsuan Yang, and Ling Shao. Multi-stage progressive image restoration. In CVPR,

  57. [65]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. In CVPR, 2022. 1, 2, 3, 4, 5

    2022

  58. [66]

    Ingredient-oriented multi- degradation learning for image restoration

    Jinghao Zhang, Jie Huang, Mingde Yao, Zizheng Yang, Hu Yu, Man Zhou, and Feng Zhao. Ingredient-oriented multi- degradation learning for image restoration. In CVPR, 2023. 1, 2, 5, 6, 7 10

    2023

  59. [67]

    Learning deep cnn denoiser prior for image restoration

    Kai Zhang, Wangmeng Zuo, Shuhang Gu, and Lei Zhang. Learning deep cnn denoiser prior for image restoration. In CVPR, 2017. 1, 2

    2017

  60. [68]

    Residual non-local attention networks for image restoration

    Yulun Zhang, Kunpeng Li, Kai Li, Bineng Zhong, and Yun Fu. Residual non-local attention networks for image restoration. arXiv preprint arXiv:1903.10082, 2019. 1, 2

    1903 arXiv

  61. [69]

    Comprehensive and delicate: An efficient transformer for image restoration

    Haiyu Zhao, Yuanbiao Gou, Boyun Li, Dezhong Peng, Jiancheng Lv, and Xi Peng. Comprehensive and delicate: An efficient transformer for image restoration. In CVPR,

  62. [70]

    Mixture-of-experts with expert choice routing

    Yanqi Zhou, Tao Lei, Hanxiao Liu, Nan Du, Yanping Huang, Vincent Zhao, Andrew M Dai, Quoc V Le, James Laudon, et al. Mixture-of-experts with expert choice routing. NeurIPS,

  63. [71]

    Learning weather- general and weather-specific features for image restoration under multiple adverse weather conditions

    Yurui Zhu, Tianyu Wang, Xueyang Fu, Xuanyu Yang, Xin Guo, Jifeng Dai, Yu Qiao, and Xiaowei Hu. Learning weather- general and weather-specific features for image restoration under multiple adverse weather conditions. In CVPR, 2023. 6

    2023

  64. [72]

    Restoring vision in adverse weather conditions with patch-based denoising diffusion models

    Ozan ¨Ozdenizci and Robert Legenstein. Restoring vision in adverse weather conditions with patch-based denoising diffusion models. TPAMI, 2023. 6 11

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.