Pith. sign in

REVIEW 3 major objections 5 minor 48 references

This paper argues that token-adaptive asymmetric attention candidate selection outperforms fixed-group attention for image super-resolution at matched compute.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-03 10:51 UTC pith:BFDVTTJW

load-bearing objection A genuinely new per-token expansion attention mechanism for SR, with good internal ablations, but the SOTA claim overreaches because baselines aren't trained under the same recipe. the 3 major comments →

arxiv 2601.08341 v2 pith:BFDVTTJW submitted 2026-01-13 cs.CV

From Local Windows to Adaptive Candidates via Individualized Exploratory: Rethinking Attention for Image Super-Resolution

Chunyu Meng , Wei Long , Shuhang Gu This is my paper
classification cs.CV
keywords single image super-resolutiontransformerattention mechanismtoken-adaptive attentionasymmetric similaritygraph expansionsparsificationindividualized exploratory attention
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that super-resolution Transformers do not need fixed groupings—windows or semantic categories—to keep attention cheap. Instead, each token can maintain its own directed list of attention candidates, chosen by content similarity, and that list can be refined across layers: nearby candidates seed the search, two-hop neighbors of high-similarity tokens are promoted to direct candidates, and low-similarity ones are pruned. The proposed IET model reports state-of-the-art PSNR/SSIM on standard SR benchmarks under comparable FLOPs to recent window- and category-based methods, with the largest gains on texture-heavy datasets like Urban100. A sympathetic reader would care because it suggests the inductive bias of fixed local windows is not necessary for efficient global attention in low-level vision.

Core claim

The central claim is that attention candidates for super-resolution should be individual, asymmetric, and content-aware rather than group-defined. Concretely, the paper introduces Individualized Exploratory Attention (IEA): each token starts with a Dense-Local Sparse-Global (DLSG) initialization—dense attention within a small region plus uniformly sampled distant tokens—then in subsequent layers expands its candidate set by gathering the top one-hop neighbors of its own top one-hop neighbors (two-hop expansion) while sparsifying away low-attention tokens. The index matrix that records each token's candidates is optimized implicitly through this expansion-and-sparsification process. The paper

What carries the argument

The central machinery is the Individualized Exploratory Attention (IEA) mechanism, built on an index matrix I ∈ R^{N×k} that explicitly stores each token's top-k attention candidates. The paper treats token similarities as a directed graph and applies two graph operations: expansion, which promotes two-hop neighbors (tokens similar to tokens that are similar to the query) into direct candidates via Eqs. 5–8, and sparsification, which prunes low-similarity candidates via top-k selection (Eq. 4). This layered expansion-and-sparsification is the mechanism that lets attention grow from local to adaptive global under a fixed computational budget.

Load-bearing premise

The load-bearing premise is that token similarity is transitive across layers: if token A is similar to B and B is similar to C in one layer, then A is similar to C in the next layer—without this transitivity, the two-hop expansion mechanism reduces to adding generic distant context rather than content-aware candidates.

What would settle it

Train the same IET architecture with expansion replaced by uniformly random long-range candidates at the same k and FLOPs; if random candidates close the 0.58 dB gap on Urban100 ×4, the two-hop transitivity assumption is not what drives the gain. Equivalently, inspect the learned attention graphs: if most two-hop promoted candidates are pruned immediately or receive near-zero attention weights, the expansion rationale is not supported.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If IET's reported gains hold, the fixed-window inductive bias in SR Transformers can be replaced by a per-token, content-driven candidate search without increasing FLOPs.
  • The ablation attributes about 0.58 dB on Urban100 ×4 to the expansion mechanism, implying that two-hop similarity propagation is the main source of the improvement over a local-only baseline.
  • IET-light matches or exceeds much larger models on several benchmarks while using a fraction of the FLOPs, suggesting the candidate-selection principle transfers to lightweight settings.
  • The paper's visualization shows attention candidates spreading beyond a 32×32 window across blocks, indicating the mechanism produces genuinely long-range content-aware receptive fields.
  • The authors state the design could transfer to other high-level vision tasks and NLP, where adaptive similarity modeling may help.

Where Pith is reading between the lines

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

  • A testable extension would measure what fraction of two-hop-expanded candidates actually survive sparsification and contribute to the output; if most are pruned or contribute little, the 0.58 dB gain might come from having more candidates rather than from the two-hop rationale.
  • The paper invokes expander-graph properties to justify DLSG initialization but does not verify that trained attention graphs satisfy the required uniqueness-of-neighborhood condition; checking the spectral gap or the overlap of one-hop neighborhoods of real models would ground or refute that justification.
  • A same-training-protocol comparison would make the reported margins directly attributable to the attention mechanism, since the baselines' published numbers come from a different optimizer and schedule.
  • The same expansion-and-sparsification loop could be applied to other token-based tasks, such as segmentation or denoising, where long-range but content-selective context is valuable—an easy extension of the code.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes the Individualized Exploratory Transformer (IET) for single-image super-resolution. IET replaces grouped attention (window-based or category-based) with an individualized exploratory attention (IEA) mechanism in which each token maintains its own asymmetric set of attention candidates. Candidates are initialized via a Dense-Local Sparse-Global (DLSG) scheme, then expanded by linking to two-hop neighbors of the most similar one-hop neighbors, and pruned by a sparsification step. A Similarity-Fused FFN further mixes features of similar tokens. Experiments on Set5, Set14, BSD100, Urban100, and Manga109 report consistent PSNR/SSIM improvements over recent window- and category-based methods at comparable FLOPs, for both a full-size model and a lightweight variant.

Significance. If the reported gains hold under controlled comparisons, IEA is a useful alternative to grouped attention for SR, and the paper's systematic evaluation across five benchmarks and two model scales is a strength. The paper also provides an inference-time comparison and visualizations of the candidate-refinement process. However, the central 'state-of-the-art under comparable computational complexity' claim is currently weakened by a training-protocol mismatch with the published baselines, and the main ablation table is misreported in the text, attributing the largest single-component gain to the wrong component. These issues need to be resolved before the headline claim can be accepted. The paper does not include machine-checked proofs; the code link is a placeholder rather than a resolved repository.

major comments (3)
  1. [§4.2, Table 1] The text and Table 1 disagree. The text states that 'the second model introduces the Expansion mechanism' and attributes about 0.58 dB improvement on Urban100 to it. In the table, the second row is DLSG only (26.75 dB) and the third row is Expansion only (26.26 dB). Thus DLSG alone gives +0.58 dB over the baseline, while Expansion alone gives only +0.09 dB. This reversal undermines the conclusion that 'directly connecting two-hop neighbors effectively expands the content-aware receptive field and brings substantial performance gains.' Please correct the text or the table and re-evaluate the role of expansion relative to DLSG in the final model.
  2. [§4.1, Supp. F, and §4.3, Tables 4–5] The headline SOTA claim is not controlled for training protocol. IET is trained with a Muon+AdamW hybrid optimizer, a two-stage schedule (50×50 then 75×75 patches, 300k+250k iterations), and a dilation change from 2 to 3, while all baselines are cited from published numbers obtained with the standard AdamW recipe. In SR, optimizer choice and patch-size schedules of this magnitude can shift PSNR by several tenths of a dB, which is exactly the range of the reported margins (e.g., 0.17 dB over PFT on Urban100 ×2; 0.33 dB for IET-light over PFT-light). Please retrain at least one strong baseline (e.g., PFT or ATD) under the identical protocol, or provide evidence that the training recipe does not affect the relative ranking.
  3. [§3.1–3.2, Eqs. (5)–(8)] The expansion mechanism rests on the transitivity premise that if A is similar to B and B is similar to C in the preceding layer, then A is likely similar to C in subsequent layers. This premise is not verified. The appeal to Expander Graphs requires a 'unique neighborhood' property, but the paper provides no measurement of whether the learned attention graphs satisfy this property, nor what fraction of expanded two-hop candidates survive sparsification and are actually used. This is not fatal to the empirical results, but it is needed to support the 'content-aware exploratory' explanation. Please add a quantitative analysis of the expanded candidates, or soften the graph-theoretic framing.
minor comments (5)
  1. [Table 5] In the lightweight SR table, the rows for MambaIRv2-light at ×3 and ×4 are labeled with Scale = ×2. The scale column should read ×3 and ×4 respectively.
  2. [Supplementary Material] The supplement text says 'in Section A ... Section B ... Section C', but the actual section headings are F, G, H. Please renumber for consistency.
  3. [Abstract] The code link is given as 'here' with no actual URL. Please provide a working repository link.
  4. [§4.3] The sentence 'IET-light outperforms SwinIR by 0.02dB on Set5 and 0.19dB on Urban100' is ambiguous: it refers to the non-light SwinIR from Table 4, not SwinIR-light from Table 5. Please clarify to avoid conflation of the two models.
  5. [§3.3] In SF-FFN, the selection of 'the most similar neighbors I_highest' is based on updated neighbor indices. It would be helpful to state whether this selection is performed per head or per token, and whether it adds meaningful computation on top of the already-computed attention indices.

Circularity Check

0 steps flagged

No significant circularity; the SOTA claim rests on external benchmark measurements, and the IEA expansion step is not defined in terms of its own output.

full rationale

The central claim is empirical: PSNR/SSIM on Set5/Set14/BSD100/Urban100/Manga109 are externally measured quantities, and IET is trained from LR-HR pairs with no fitted parameter that is then reported as the target metric. Eq. (2)-(3) define individualized attention using an index matrix I, and Eqs. (4)-(8) specify sparsification/expansion as a graph operation on previous-layer attention maps; the final PSNR is not an algebraic consequence of those equations. The ablation in Table 1 changes one component at a time under a fixed protocol, so the 0.58 dB expansion gain is an internally controlled comparison, not a fitted prediction. Self-citations exist (PFT [27] and ATD [44] share authors; the CUDA SMM framework and progressive attention are taken from [27]), but accepting PFT's or ATD's conclusions is not needed for IET's reported numbers; the citations are implementation scaffolding and baseline sources. No uniqueness theorem or ansatz is imported from the authors' prior work to force the design. The expander-graph premise in Sec. 3.2 ('this property holds if each node has a unique neighborhood') is asserted without verification, and the Muon-vs-AdamW protocol mismatch with baselines is a potential comparison-fairness concern; both are correctness risks, not circularity. Hence score 2 rather than higher.

Axiom & Free-Parameter Ledger

5 free parameters · 3 axioms · 0 invented entities

No new physical or external entities are postulated. The IEA index matrix I is the method's own data structure, not an unexplained assumption. The free parameters are the hand-chosen candidate-count hyperparameters and dilation settings that the central benchmark claim depends on; the axioms are the transitivity heuristic, the expander-graph assertion, and the comparability of differently trained baselines.

free parameters (5)
  • k1 (one-hop expansion counts) = [22, 20, 14, 12]
    Section 3.4 - hand-chosen per-block counts of one-hop neighbors selected for expansion; no sensitivity study reported.
  • k2 (two-hop neighbor counts) = [12, 11, 9, 8]
    Section 3.4 - hand-chosen expansion breadth per block.
  • DLSG dilation d = 2 (training) / 3 (inference)
    Section 4.2, Table 2 - selected by training/inference ablation; different values used at train and test time.
  • Number of expansion steps = 4
    Section 4.2, Table 3 - ablation shows 4 beats 2, 3, and 5.
  • Sparsification top-K k_s = not reported
    Eq. 4 - the sparsification budget k_s is never given a numeric value in the paper, yet it determines how aggressively candidates are pruned.
axioms (3)
  • domain assumption Two-hop similarity transitivity: if A is similar to B and B is similar to C in the preceding layer, then A is likely similar to C in subsequent layers.
    Sections 3.1-3.2 - the premise of the expansion mechanism (Eqs. 5-7); supported empirically by Table 1 but never verified as a property of the trained attention graphs.
  • domain assumption Expander-graph uniqueness: the DLSG initialization guarantees each token discovers two-hop neighbors distinct from its one-hop ones.
    Section 3.2 - asserted ('this property holds if each node has a unique neighborhood') rather than proven for the actual d x d uniform sampling.
  • domain assumption Published baseline PSNR/SSIM numbers are comparable despite different training protocols (Muon vs AdamW, patch sizes, iteration counts).
    Section 4.1 vs Tables 4-5 - required for the 'state-of-the-art under comparable computational complexity' claim; no baseline is retrained under the IET protocol.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of From Local Windows to Adaptive Candidates via Individualized Exploratory: Rethinking Attention for Image Super-Resolution." pith.science (2026). https://pith.science/paper/BFDVTTJW

@misc{pith2026260108341,
  author       = {Pith},
  title        = {Pith review of: From Local Windows to Adaptive Candidates via Individualized Exploratory: Rethinking Attention for Image Super-Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BFDVTTJW}},
  note         = {Machine review of arXiv:2601.08341}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Single Image Super-Resolution (SISR) is a fundamental computer vision task that aims to reconstruct a high-resolution (HR) image from a low-resolution (LR) input. Transformer-based methods have achieved remarkable performance by modeling long-range dependencies in degraded images. However, their feature-intensive attention computation incurs high computational cost. To improve efficiency, most existing approaches partition images into fixed groups and restrict attention within each group. Such group-wise attention overlooks the inherent asymmetry in token similarities, thereby failing to enable flexible and token-adaptive attention computation. To address this limitation, we propose the Individualized Exploratory Transformer (IET), which introduces a novel Individualized Exploratory Attention (IEA) mechanism that allows each token to adaptively select its own content-aware and independent attention candidates. This token-adaptive and asymmetric design enables more precise information aggregation while maintaining computational efficiency. Extensive experiments on standard SR benchmarks demonstrate that IET achieves state-of-the-art performance under comparable computational complexity.

Figures

Figures reproduced from arXiv: 2601.08341 by Chunyu Meng, Shuhang Gu, Wei Long.

Figure 1
Figure 1. Figure 1: Comparison of different attention mechanism. (a) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed DLSG initialization strategy. In the nearby [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The proposed sparsification and expansion mechanism. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: The proposed Similarity-Fused Feed-Forward Network [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual comparisons of IET and other state-of-the-art image super-resolution methods. Query Token Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of the attention candidates initialization and expansion process. The yellow grid represents a 32×32 window, which [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of the attention candidates initialization and expansion process. The yellow grid represents a 32×32 window, which [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual comparisons of IET and other state-of-the-art image super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visual comparisons of IET and other state-of-the-art image super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visual comparisons of IET-light and other state-of-the-art image super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visual comparisons of IET-light and other state-of-the-art image super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

48 extracted references · 3 linked inside Pith

  1. [1]

    Fast, accurate, and lightweight super-resolution with cascading residual network, 2018

    Namhyuk Ahn, Byungkon Kang, and Kyung-Ah Sohn. Fast, accurate, and lightweight super-resolution with cascading residual network, 2018. 7, 8

    2018

  2. [2]

    Low-complexity single-image super-resolution based on nonnegative neighbor embedding

    Marco Bevilacqua, Aline Roumy, Christine Guillemot, and Marie-line Alberi Morel. Low-complexity single-image super-resolution based on nonnegative neighbor embedding. InProcedings of the British Machine Vision Conference 2012, 2012. 7

    2012

  3. [3]

    Pre-trained image processing transformer, 2020

    Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer, 2020. 2, 6

    2020

  4. [4]

    Activating more pixels in image super- resolution transformer

    Xiangyu Chen, Xintao Wang, Jiantao Zhou, Yu Qiao, and Chao Dong. Activating more pixels in image super- resolution transformer. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 22367–22377, 2023. 1, 2, 5, 6, 8

    2023

  5. [5]

    Cross aggregation transformer for image restoration

    Zheng Chen, Yulun Zhang, Jinjin Gu, Yongbing Zhang, Linghe Kong, and Xin Yuan. Cross aggregation transformer for image restoration. InNeurIPS, 2022. 1, 2, 6, 8

    2022

  6. [6]

    N-gram in swin transformers for efficient lightweight image super- resolution, 2022

    Haram Choi, Jeongmin Lee, and Jihoon Yang. N-gram in swin transformers for efficient lightweight image super- resolution, 2022. 7

    2022

  7. [7]

    Twins: Revisiting the design of spatial attention in vision transformers, 2021

    Xiangxiang Chu, Zhi Tian, Yuqing Wang, Bo Zhang, Haib- ing Ren, Xiaolin Wei, Huaxia Xia, and Chunhua Shen. Twins: Revisiting the design of spatial attention in vision transformers, 2021. 2

    2021

  8. [8]

    Second-order attention network for single image super-resolution

    Tao Dai, Jianrui Cai, Yongbing Zhang, Shu-Tao Xia, and Lei Zhang. Second-order attention network for single image super-resolution. In2019 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2020. 2

    2020

  9. [9]

    Learning a deep convolutional network for image super-resolution

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Learning a deep convolutional network for image super-resolution. InComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part IV, pages 184–199. Springer, 2014. 1

    2014

  10. [10]

    Image super-resolution using deep convolutional net- works.IEEE Transactions on Pattern Analysis and Machine Intelligence, page 295–307, 2015

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works.IEEE Transactions on Pattern Analysis and Machine Intelligence, page 295–307, 2015. 1, 2

    2015

  11. [11]

    Cswin transformer: A general vision transformer backbone with cross-shaped windows, 2021

    Xiaoyi Dong, Jianmin Bao, Dongdong Chen, Weiming Zhang, Nenghai Yu, Lu Yuan, Dong Chen, and Baining Guo. Cswin transformer: A general vision transformer backbone with cross-shaped windows, 2021. 5

    2021

  12. [12]

    An image is worth 16x16 words: Transformers for image recognition at scale, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale, 2020. 2

    2020

  13. [13]

    Convolutional sparse coding for image super-resolution

    Shuhang Gu, Wangmeng Zuo, Qi Xie, Deyu Meng, Xi- angchu Feng, and Lei Zhang. Convolutional sparse coding for image super-resolution. InProceedings of the IEEE Inter- national Conference on Computer Vision, pages 1823–1831,

  14. [14]

    Learned dynamic guidance for depth image reconstruction.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(10):2437– 2452, 2019

    Shuhang Gu, Shi Guo, Wangmeng Zuo, Yunjin Chen, Radu Timofte, Luc Van Gool, and Lei Zhang. Learned dynamic guidance for depth image reconstruction.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(10):2437– 2452, 2019. 2

    2019

  15. [15]

    Mambairv2: Attentive state space restoration.arXiv preprint arXiv:2411.15269,

    Hang Guo, Yong Guo, Yaohua Zha, Yulun Zhang, Wenbo Li, Tao Dai, Shu-Tao Xia, and Yawei Li. Mambairv2: Attentive state space restoration.arXiv preprint arXiv:2411.15269,

    Pith/arXiv arXiv

  16. [16]

    Sin- gle image super-resolution from transformed self-exemplars

    Jia-Bin Huang, Abhishek Singh, and Narendra Ahuja. Sin- gle image super-resolution from transformed self-exemplars. In2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015. 7

    2015

  17. [17]

    Lightweight image super-resolution with information multi- distillation network

    Zheng Hui, Xinbo Gao, Yunchu Yang, and Xiumei Wang. Lightweight image super-resolution with information multi- distillation network. InProceedings of the 27th ACM Inter- national Conference on Multimedia, 2019. 7, 8

    2019

  18. [18]

    Muon: An optimizer for hidden layers in neural networks, 2024

    Keller Jordan, Yuchen Jin, Vlado Boza, You Jiacheng, Franz Cesista, Laker Newhouse, and Jeremy Bernstein. Muon: An optimizer for hidden layers in neural networks, 2024. 5, 1

    2024

  19. [19]

    Accurate image super-resolution using very deep convolutional net- works

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Accurate image super-resolution using very deep convolutional net- works. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. 1, 2

    2016

  20. [20]

    Deeply- recursive convolutional network for image super-resolution

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Deeply- recursive convolutional network for image super-resolution. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1637–1645, 2016. 2

    2016

  21. [21]

    Lapar: Linearly-assembled pixel-adaptive re- gression network for single image super-resolution and be- yond, 2020

    Wenbo Li, Kun Zhou, Lu Qi, Nianjuan Jiang, Jiangbo Lu, and Jiaya Jia. Lapar: Linearly-assembled pixel-adaptive re- gression network for single image super-resolution and be- yond, 2020. 7, 8

    2020

  22. [22]

    Localvit: Bringing locality to vision transformers

    Yawei Li, Kai Zhang, Jiezhang Cao, Radu Timofte, and Luc Van Gool. Localvit: Bringing locality to vision transformers. arXiv preprint arXiv:2104.05707, 2021. 5

    Pith/arXiv arXiv 2021

  23. [23]

    Swinir: Image restora- tion using swin transformer, 2021

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restora- tion using swin transformer, 2021. 1, 2, 5, 6, 7, 8

    2021

  24. [24]

    Enhanced deep residual networks for single image super-resolution

    Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. Enhanced deep residual networks for single image super-resolution. In2017 IEEE Conference on Com- puter Vision and Pattern Recognition Workshops (CVPRW),

  25. [25]

    Catanet: Ef- ficient content-aware token aggregation for lightweight im- age super-resolution

    Xin Liu, Jie Liu, Jie Tang, and Gangshan Wu. Catanet: Ef- ficient content-aware token aggregation for lightweight im- age super-resolution. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 17902–17912, 2025. 1, 2

    2025

  26. [26]

    Swin trans- former: Hierarchical vision transformer using shifted win- dows, 2021

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin trans- former: Hierarchical vision transformer using shifted win- dows, 2021. 1, 2

    2021

  27. [27]

    Progressive focused transformer for single image super- resolution.arXiv preprint arXiv:2503.20337, 2025

    Wei Long, Xingyu Zhou, Leheng Zhang, and Shuhang Gu. Progressive focused transformer for single image super- resolution.arXiv preprint arXiv:2503.20337, 2025. 1, 2, 5, 6, 7, 8

    Pith/arXiv arXiv 2025

  28. [28]

    Decoupled weight de- cay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight de- cay regularization. InInternational Conference on Learning Representations, 2018. 1

    2018

  29. [29]

    Xiaotong Luo, Yuan Xie, Yulun Zhang, Yanyun Qu, Cui- hua Li, and Yun Fu.LatticeNet: Towards Lightweight Image Super-Resolution with Lattice Block, page 272–289. 2020. 7

    2020

  30. [30]

    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 ecolog- ical statistics. InProceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001, 2002. 7

    2001

  31. [31]

    Sketch-based manga retrieval using manga109 dataset.Mul- timedia Tools and Applications, page 21811–21838, 2016

    Yusuke Matsui, Kota Ito, Yuji Aramaki, Azuma Fujimoto, Toru Ogawa, Toshihiko Yamasaki, and Kiyoharu Aizawa. Sketch-based manga retrieval using manga109 dataset.Mul- timedia Tools and Applications, page 21811–21838, 2016. 7

    2016

  32. [32]

    Image super- resolution with cross-scale non-local attention and exhaus- tive self-exemplars mining

    Yiqun Mei, Yuchen Fan, Yuqian Zhou, Lichao Huang, Thomas S Huang, and Humphrey Shi. Image super- resolution with cross-scale non-local attention and exhaus- tive self-exemplars mining. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020. 2

    2020

  33. [33]

    Image super-resolution with non-local sparse attention

    Yiqun Mei, Yuchen Fan, and Yuqian Zhou. Image super-resolution with non-local sparse attention. In2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 2

    2021

  34. [34]

    Ben Niu, Weilei Wen, Wenqi Ren, Xiangde Zhang, Lianping Yang, Shuzhen Wang, Kaihao Zhang, Xiaochun Cao, and Haifeng Shen.Single Image Super-Resolution via a Holistic Attention Network, page 191–207. 2020. 2, 6, 8

    2020

  35. [35]

    Image processing gnn: Breaking rigidity in super-resolution

    Yuchuan Tian, Hanting Chen, Chao Xu, and Yunhe Wang. Image processing gnn: Breaking rigidity in super-resolution. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 24108– 24117, 2024. 3, 6, 7, 8, 1

    2024

  36. [36]

    Ntire 2017 chal- lenge on single image super-resolution: Methods and results

    Radu Timofte, Eirikur Agustsson, Luc Van Gool, Ming- Hsuan Yang, Lei Zhang, Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, Kyoung Mu Lee, et al. Ntire 2017 chal- lenge on single image super-resolution: Methods and results. In2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2017. 5, 1

    2017

  37. [37]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need, 2017. 2, 3

    2017

  38. [38]

    Omni aggregation networks for lightweight im- age super-resolution

    Hang Wang, Xuanhong Chen, Bingbing Ni, Yutian Liu, and Liu jinfan. Omni aggregation networks for lightweight im- age super-resolution. InConference on Computer Vision and Pattern Recognition, 2023. 5, 7, 8

    2023

  39. [39]

    Pyra- mid vision transformer: A versatile backbone for dense pre- diction without convolutions

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyra- mid vision transformer: A versatile backbone for dense pre- diction without convolutions. In2021 IEEE/CVF Interna- tional Conference on Computer Vision (ICCV), 2022. 2

    2022

  40. [40]

    Non-local neural networks

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaim- ing He. Non-local neural networks. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 7794–7803, 2018. 2

    2018

  41. [41]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 17683–17693, 2022. 5

    2022

  42. [42]

    Roman Zeyde, Michael Elad, and Matan Protter.On Sin- gle Image Scale-Up Using Sparse-Representations, page 711–730. 2012. 7

    2012

  43. [43]

    Accurate image restoration with attention retractable transformer

    Jiale Zhang, Yulun Zhang, Jinjin Gu, Yongbing Zhang, Linghe Kong, and Xin Yuan. Accurate image restoration with attention retractable transformer. InICLR, 2023. 6, 8

    2023

  44. [44]

    Transcending the limit of local window: Ad- vanced super-resolution transformer with adaptive token dic- tionary

    Leheng Zhang, Yawei Li, Xingyu Zhou, Xiaorui Zhao, and Shuhang Gu. Transcending the limit of local window: Ad- vanced super-resolution transformer with adaptive token dic- tionary. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2856–2865, 2024. 1, 2, 5, 6, 7, 8

    2024

  45. [45]

    Efficient long-range attention network for image super- resolution

    Xindong Zhang, Hui Zeng, Shi Guo, and Lei Zhang. Efficient long-range attention network for image super- resolution. InEuropean Conference on Computer Vision, pages 649–667. Springer, 2022. 1, 7, 8

    2022

  46. [46]

    Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu.Image Super-Resolution Using Very Deep Residual Channel Attention Networks, page 294–310

  47. [47]

    Residual dense network for image super-resolution,

    Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, and Yun Fu. Residual dense network for image super-resolution,

  48. [48]

    Srformer: Permuted self- attention for single image super-resolution

    Yupeng Zhou, Zhen Li, Chun-Le Guo, Song Bai, Ming- Ming Cheng, and Qibin Hou. Srformer: Permuted self- attention for single image super-resolution. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 12780–12791, 2023. 1 From Local Windows to Adaptive Candidates via Individualized Exploratory: Rethinking Attention ...

    2023

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.