Pith. sign in

REVIEW 5 major objections 4 minor 73 references

A Tree-guided CNN for image super-resolution

T0 review · 5 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper proposes TSRNet, a convolutional super-resolution network whose four feature-extraction branches are fused in a binary-tree order, and reports it outperforms popular baselines on four standard benchmarks while being faster and…

desk verdict A competent but incremental multi-branch SR architecture whose 'tree-guide' claim is not isolated by the experiments; worth a referee, but only with heavy revision. read the letter →

arxiv 2506.02585 v1 pith:XHNGFPYP submitted 2025-06-03 eess.IV cs.CV

Chunwei Tian , Mingjian Song , Xiaopeng Fan , Xiangtao Zheng , Bob Zhang , David Zhang This is my paper
classification eess.IVcs.CV
keywords imagesuper-resolutiontreenetworkarchitecturedeepCNNcosinetransformAdanoptimizerhierarchicalfeaturefusionlightweight
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

TSRNet is a convolutional network for single-image super-resolution that organizes four feature-extracting branches into a binary-tree structure, with outputs merged progressively by three fusion blocks in the order ((T1,T2),T3),T4. The paper claims this tree guidance, combined with cosine-transform blocks that extract directional local structure and the Adan optimizer that stabilizes training, yields higher PSNR and SSIM than popular SR methods on Set5, Set14, BSD100, and Urban100 at x2, x3, and x4 upscaling. The model is also reported to be faster and lighter than several strong baselines. If the claim holds, it suggests that explicitly structuring a network as a tree of branches is a cheap way to improve super-resolution quality without scaling up model size.

What carries the argument

The central mechanism is the binary-tree branch topology: four feature-extraction branches are fused pairwise in a fixed hierarchy, with each fusion block (a residual addition followed by five convolutional blocks) refining the merged features. Each branch is a stack of nine blocks of 3x3 convolution plus ReLU; the first two branches and the first fusion block also contain a cosine-transform mechanism block (CTMB), which uses sharpened cosine similarity instead of a plain dot product to extract directional local features. Training uses Adan, an adaptive Nesterov-momentum optimizer, to suppress gradient explosion and stabilize deep-network training.

What would settle it

Train a super-resolution network with the same number of parameters and layers as TSRNet but with all four branches merged in a single parallel step rather than the tree order, using the same DIV2K training data and Adan settings; if its PSNR on Urban100 x4 matches or exceeds TSRNet's 26.00 dB, the tree ordering is not the cause of the reported gains.

Watch

Extended reading notes

Core claim

The paper's central claim is that TSRNet produces higher-quality super-resolution images than established methods while staying efficient. The architecture uses four branches of stacked 3x3 convolutions and ReLUs, with the first two branches and the first fusion block also containing cosine-transform blocks. A chain of three fusion blocks merges the branches one pair at a time, and the paper argues this tree ordering lets the network enhance the relation of key nodes so that important layers are not drowned out by unimportant ones. Its ablation on Urban100 x4 shows each added component, from the second pair of branches through the cosine blocks to the Adan optimizer, raises PSNR. Full comparisons on four benchmarks report TSRNet above all listed baselines, such as exceeding AFAN-S by 0.08 dB at x2 on Set5 and DSRCNN by 0.09 dB at x3 on Urban100.

Load-bearing premise

The load-bearing premise is that the binary-tree fusion topology itself, rather than the extra depth, the cosine blocks, or the Adan optimizer, is what improves super-resolution; the paper never tests this premise directly.

Editorial extensions

If this is right

  • TSRNet is reported to achieve higher PSNR and SSIM than over twenty listed methods on Set5, Set14, BSD100, and Urban100 at x2, x3, and x4, so the tree hierarchy plus cosine blocks and Adan form an effective recipe for super-resolution.
  • The model is efficient: about 2.25 million parameters and 181.25 GFLOPs for x4 super-resolution on 1024x1024 inputs, compared with 13.6 million parameters and 498.18 GFLOPs for DCLS, and lower running time than EDSR, CARN-M, and ACNet.
  • The ablation on Urban100 x4 attributes measurable PSNR gains to each design choice, including adding the third and fourth branches, adding cosine-transform blocks, and switching the optimizer from Adam to Adan.
  • Because the tree fusions are simple residual additions, the design is implementation-friendly and could be ported to other image restoration tasks.
  • The reported margin over the strongest baselines is small in absolute terms (0.06 to 0.09 dB), which is typical for ranking super-resolution methods in this benchmark tradition.

Reading between the lines

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

  • The tree fusion order is one design point in a space of possible merge hierarchies; extending the method to search over such orders, or to learn them, is a natural follow-up that the paper's formula (1) makes easy to define.
  • Because the cosine-transform blocks are the only component operating on directional similarity, TSRNet's margin over non-cosine baselines should be most visible on edge-rich images; a per-image breakdown of the Urban100 results would show whether that prediction holds.
  • Adan is a general-purpose optimizer, so the reported gain from switching Adam to Adan likely transfers to other super-resolution backbones; re-training a published baseline with Adan would test that transfer.
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

5 major / 4 minor

Summary. The paper proposes TSRNet, a CNN for single-image super-resolution that combines four parallel branches (called tree branches) with fusion blocks, a cosine transform mechanism block, and the Adan optimizer. The authors claim that the tree architecture enhances hierarchical information and that cosine convolution extracts cross-domain salient features, yielding superior PSNR/SSIM on Set5, Set14, B100, and U100 for x2, x3, and x4, along with competitive complexity and runtime.

Significance. If the tree-guidance hypothesis were convincingly validated, the paper would offer a modest but useful design insight for lightweight SR networks, with a public code release and complexity numbers (2,253K parameters, 181.25G FLOPs) that are attractive for consumer-electronics applications. The empirical protocol, however, does not currently isolate the tree topology from changes in model capacity, the cosine block, or the optimizer, and the reported PSNR gains are small point estimates without repeated runs. The architectural and experimental evidence therefore does not yet support the central 'superiority' claim, though the issues are addressable within the manuscript's scope.

major comments (5)
  1. [§IV.C, Table I] The ablation study removes several components at once (CTMB, Adan, and branches T1-T3) and is reported only on U100 ×4. Because parameter count, optimizer, and architecture change simultaneously, the PSNR differences in Table I (25.19 to 26.00 dB) do not isolate the contribution of the tree topology. A matched-capacity control with the same number of parameters and training schedule—for example, a plain multi-branch or sequential residual network without the tree-specific fusion—is needed before the observed gains can be attributed to tree guidance.
  2. [§IV.D, Tables II-V] The claimed superiority rests on point estimates with margins of 0.07 to 0.09 dB in most cases (e.g., TSRNet 37.92 vs AFAN-S 37.84 on Set5 ×2, 32.15 vs MemNet 32.08 on B100 ×2, 28.08 vs DSRCNN 27.99 on U100 ×3). No standard deviations, repeated runs, or significance tests are reported, and differences of this magnitude are within seed-level training noise in super-resolution. Moreover, TSRNet is not consistently the best method across the tables (e.g., on Set5 ×4 it is below DCLS and AFAN-S), so the conclusion that TSRNet is 'superior to popular SR methods' is not supported by the reported evidence.
  3. [§IV.D, Table III] Table III lists 'CRFAN 8.35/0.7790' for Set14 ×4; a PSNR of 8.35 dB is not a plausible value for this benchmark and appears to be a typographical error (likely 28.35). Because the quantitative tables are the primary evidence for the paper's central claim, such errors must be corrected and the numerical results re-verified before the comparisons can be considered reliable.
  4. [§III.A, Eqs. (2)-(3)] The architecture description is ambiguous: Eqs. (2)-(3) equate 9BTBB(R(Conv(ILR))) with 9R(Conv(R(Conv(ILR)))), which does not clarify how a BTBB is stacked, and they also state T1=T2 and T3=T4, so the four 'tree branches' reduce to only two distinct transformations. This notation obscures how the structure differs from a parallel multi-branch residual network, and the fusion of two identical signals in Eq. (4) suggests the tree topology may not be essential. The authors should clarify the BTBB definition and explain how a true tree structure, rather than replicated branches, is obtained.
  5. [§IV.D, comparison list] The comparison list omits widely used recent SR models such as RCAN, SwinIR, and HAT, so the statement that TSRNet is 'superior to popular SR methods' is not tested against the current state of the art. The authors should either include such baselines or explicitly restrict the claim to the methods compared in Tables II-V.
minor comments (4)
  1. [§III.C, Eq. (10) and text] The phrase 'That at, firstly' is ungrammatical, and the learnable parameters ε and p in Eq. (10) are not described in terms of initialization or training behavior, which makes the cosine convolution mechanism difficult to reproduce.
  2. [§IV.D and references] Several formatting errors appear, such as 'FDSR [55]]' and 'CSCN [51]]' in the text, and the reference to the cosine convolution source [36] is a blog post rather than a peer-reviewed publication; these should be corrected.
  3. [§III.A, Eq. (7)] The text says 'P R denotes a function of a pixel-shuffle function' but the equation uses PS; the notation should be made consistent.
  4. [§IV.C, Table I caption] The caption 'MEANPSNRRESULTS FOR DIFFERENT METHODS WITH ×4 ON U100' is missing spaces and the table reports only PSNR, not SSIM; the ablation variant names should be defined explicitly for readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's PSNR claims are self-contained empirical measurements against external baselines, and its architecture is defined by explicit equations rather than by the results it reports.

full rationale

The paper's derivation chain is descriptive rather than inferential: Eqs. (1)-(7) define the TSRNet architecture, Eq. (8) is the standard MSE objective, Eqs. (9)-(10) define the CTMB from the externally cited cosine convolution [36], and Eqs. (11)-(15) restate the Adan optimizer from its external source [34]. The performance claims are direct experimental measurements on Set5, Set14, B100, and U100 against published external baselines; no PSNR or SSIM value is obtained by fitting a parameter and then re-predicting that same quantity. Table I is a standard component-removal ablation and, while it confounds capacity and is run on a single dataset, that is an experimental-validity concern rather than a circular derivation. The paper cites several prior works by the same authors (e.g., [35], [45], [61]), but these are used as general design motivation and not as a uniqueness theorem or as the sole justification for the central empirical claim. Thus no load-bearing step reduces to its own inputs by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central result is an empirical architecture; it rests on many hand-chosen hyperparameters and standard domain assumptions about SR evaluation. No new theoretical entities are introduced.

free parameters (6)
  • Number of tree branches = 4
    The architecture uses four branches T1-T4, chosen by hand; no evidence that four is optimal.
  • BTBB depth per branch = 9
    Each branch stacks nine BTBB blocks (conv+ReLU); chosen empirically without sensitivity analysis.
  • Fusion block BTBB depth = 5
    Each of the three fusion blocks uses five BTBB blocks; chosen by hand.
  • Initial learning rate = 4e-4
    Set manually, with multi-step decay by half every 200 epochs.
  • Adan optimizer hyperparameters = beta1=0.98, beta2=0.92, beta3=0.99, epsilon=1e-8
    Taken from the Adan paper but not justified for this model.
  • Cosine convolution learnable parameters = epsilon and p in Eq. (10)
    Learnable parameters controlling the sharpened cosine similarity; introduced as part of the SCS formulation.
assumptions (5)
  • domain assumption MSE loss is an appropriate training objective for SR
    The paper uses MSE without discussion of perceptual quality trade-offs, a standard but debatable choice.
  • domain assumption PSNR/SSIM on standard benchmarks measure SR quality
    These metrics are standard but imperfect proxies for visual quality; the paper treats them as definitive.
  • domain assumption Cosine convolution from a blog post is valid and reliable
    Ref. [36] is a personal blog post, not peer-reviewed; the paper adopts it as a building block without independent validation.
  • domain assumption Adan optimizer improves SR training over Adam
    The paper relies on Ref. [34] for Adan; the claimed improvement is assumed to transfer to this architecture.
  • domain assumption Training on DIV2K generalizes to the test datasets
    Standard practice, but no validation set is used for early stopping or model selection.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Tree-guided CNN for image super-resolution." pith.science (2026). https://pith.science/paper/XHNGFPYP

@misc{pith2026250602585,
  author       = {Pith},
  title        = {Pith review of: A Tree-guided CNN for image super-resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XHNGFPYP}},
  note         = {Machine review of arXiv:2506.02585}
}
read the original abstract

Deep convolutional neural networks can extract more accurate structural information via deep architectures to obtain good performance in image super-resolution. However, it is not easy to find effect of important layers in a single network architecture to decrease performance of super-resolution. In this paper, we design a tree-guided CNN for image super-resolution (TSRNet). It uses a tree architecture to guide a deep network to enhance effect of key nodes to amplify the relation of hierarchical information for improving the ability of recovering images. To prevent insufficiency of the obtained structural information, cosine transform techniques in the TSRNet are used to extract cross-domain information to improve the performance of image super-resolution. Adaptive Nesterov momentum optimizer (Adan) is applied to optimize parameters to boost effectiveness of training a super-resolution model. Extended experiments can verify superiority of the proposed TSRNet for restoring high-quality images. Its code can be obtained at https://github.com/hellloxiaotian/TSRNet.

Figures

Figures reproduced from arXiv: 2506.02585 by the authors.

Figure 1
Figure 1. Network architecture of the proposed TSRNet. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Comparisons of visual results of different super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Comparisons of visual results of different super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Comparisons of visual results of different super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: Comparisons of visual results of different super-resolution methods [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

73 extracted references · 54 canonical work pages

  1. [1]

    Deep learning for single image super-resolution: A brief review,

    W. Yang, X. Zhang, Y . Tian, W. Wang, J.-H. Xue, and Q. Liao, “Deep learning for single image super-resolution: A brief review,”IEEE Transactions on Multimedia, vol. 21, no. 12, pp. 3106–3121, 2019

    work page 2019

  2. [2]

    Nearest neighbor value interpolation,

    O. Rukundo and H. Cao, “Nearest neighbor value interpolation,”arXiv preprint arXiv:1211.1768, 2012

    arXiv 2012

  3. [3]

    New edge-directed interpolation,

    X. Li and M. T. Orchard, “New edge-directed interpolation,”IEEE transactions on image processing, vol. 10, no. 10, pp. 1521–1527, 2001

    work page 2001

  4. [4]

    Cubic convolution interpolation for digital image processing,

    R. Keys, “Cubic convolution interpolation for digital image processing,” IEEE transactions on acoustics, speech, and signal processing, vol. 29, no. 6, pp. 1153–1160, 1981

    work page 1981

  5. [5]

    Super-resolution image reconstruction: a technical overview,

    S. C. Park, M. K. Park, and M. G. Kang, “Super-resolution image reconstruction: a technical overview,”IEEE signal processing magazine, vol. 20, no. 3, pp. 21–36, 2003

    work page 2003

  6. [6]

    Minimax concave penalty regression for superresolution image reconstruction,

    X. Liao, X. Wei, and M. Zhou, “Minimax concave penalty regression for superresolution image reconstruction,”IEEE Transactions on Consumer Electronics, vol. 70, no. 1, pp. 2999-3007, 2023

    work page 2023

  7. [7]

    Image super-resolution via sparse representation,

    J. Yang, J. Wright, T. S. Huang, and Y . Ma, “Image super-resolution via sparse representation,”IEEE transactions on image processing, vol. 19, no. 11, pp. 2861–2873, 2010

    work page 2010

  8. [8]

    Image super-resolution as sparse representation of raw image patches,

    J. Yang, J. Wright, T. Huang, and Y . Ma, “Image super-resolution as sparse representation of raw image patches,” in2008 IEEE conference on computer vision and pattern recognition. IEEE, 2008, pp. 1–8

    work page 2008

Show all 73 references
  1. [9]

    A non-local approach for image super-resolution using intermodality priors,

    F. Rousseau, A. D. N. Initiativeet al., “A non-local approach for image super-resolution using intermodality priors,”Medical image analysis, vol. 14, no. 4, pp. 594–605, 2010

    2010

  2. [10]

    Learning a deep convolutional network for image super-resolution,

    C. Dong, C. C. Loy, K. He, and X. 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 13. Springer, 2014, pp. 184–199

    2014

  3. [11]

    Coarse-to- fine cnn for image super-resolution,

    C. Tian, Y . Xu, W. Zuo, B. Zhang, L. Fei, and C.-W. Lin, “Coarse-to- fine cnn for image super-resolution,”IEEE Transactions on Multimedia, vol. 23, pp. 1489–1502, 2020

    2020

  4. [12]

    Image super-resolution using deep convolutional networks,

    C. Dong, C. C. Loy, K. He, and X. Tang, “Image super-resolution using deep convolutional networks,”IEEE transactions on pattern analysis and machine intelligence, vol. 38, no. 2, pp. 295–307, 2015

    2015

  5. [13]

    Runet: A robust unet architecture for image super-resolution,

    X. Hu, M. A. Naiel, A. Wong, M. Lamm, and P. Fieguth, “Runet: A robust unet architecture for image super-resolution,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019, pp. 0–0

    2019

  6. [14]

    Accurate image super-resolution using very deep convolutional networks,

    J. Kim, J. K. Lee, and K. M. Lee, “Accurate image super-resolution using very deep convolutional networks,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 1646– 1654

    2016

  7. [15]

    Deeply-recursive convolutional net- work for image super-resolution,

    J. Kim, J. K. Lee, and K. M. Lee, “Deeply-recursive convolutional net- work for image super-resolution,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 1637–1645

    2016

  8. [16]

    Accelerating the super-resolution convolutional neural network,

    C. Dong, C. C. Loy, and X. Tang, “Accelerating the super-resolution convolutional neural network,” inComputer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14. Springer, 2016, pp. 391–407

    2016

  9. [17]

    Asymmetric cnn for image superresolution,

    C. Tian, Y . Xu, W. Zuo, C.-W. Lin, and D. Zhang, “Asymmetric cnn for image superresolution,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 52, no. 6, pp. 3718–3730, 2021

    2021

  10. [18]

    Hybrid attention feature refinement network for lightweight image super-resolution in metaverse immersive display,

    K. Wanga, X. Yanga, and G. Jeon, “Hybrid attention feature refinement network for lightweight image super-resolution in metaverse immersive display,”IEEE Transactions on Consumer Electronics, vol. 70, no. 1, pp. 3232–3244, 2023

    2023

  11. [19]

    Compressed multi- scale feature fusion network for single image super-resolution,

    X. Fan, Y . Yang, C. Deng, J. Xu, and X. Gao, “Compressed multi- scale feature fusion network for single image super-resolution,”Signal processing, vol. 146, pp. 50–60, 2018

    2018

  12. [20]

    Memnet: A persistent memory network for image restoration,

    Y . Tai, J. Yang, X. Liu, and C. Xu, “Memnet: A persistent memory network for image restoration,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 4539–4547

    2017

  13. [21]

    Image super-resolution via deep recursive residual network,

    Y . Tai, J. Yang, and X. Liu, “Image super-resolution via deep recursive residual network,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 3147–3155

    2017

  14. [22]

    Deep learning for image super- resolution: A survey,

    Z. Wang, J. Chen, and S. C. Hoi, “Deep learning for image super- resolution: A survey,”IEEE transactions on pattern analysis and ma- chine intelligence, vol. 43, no. 10, pp. 3365–3387, 2020

    2020

  15. [23]

    Image super-resolution using dense skip connections,

    T. Tong, G. Li, X. Liu, and Q. Gao, “Image super-resolution using dense skip connections,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 4799–4807

    2017

  16. [24]

    Lightweight image super- resolution with information multi-distillation network,

    Z. Hui, X. Gao, Y . Yang, and X. Wang, “Lightweight image super- resolution with information multi-distillation network,” inProceedings of the 27th acm international conference on multimedia, 2019, pp. 2024– 2032

    2019

  17. [25]

    Photo-realistic single image super-resolution using a generative adversarial network,

    C. Ledig, L. Theis, F. Husz ´ar, J. Caballero, A. Cunningham, A. Acosta, A. Aitken, A. Tejani, J. Totz, Z. Wanget al., “Photo-realistic single image super-resolution using a generative adversarial network,” in Proceedings of the IEEE conference on computer vision and pattern r...

    2017

  18. [26]

    Ctcnet: A cnn- transformer cooperation network for face image super-resolution,

    G. Gao, Z. Xu, J. Li, J. Yang, T. Zeng, and G.-J. Qi, “Ctcnet: A cnn- transformer cooperation network for face image super-resolution,”IEEE Transactions on Image Processing, vol. 32, pp. 1978–1991, 2023

    1978

  19. [27]

    Super-fan: Integrated facial landmark localization and super-resolution of real-world low resolution faces in arbitrary poses with gans,

    A. Bulat and G. Tzimiropoulos, “Super-fan: Integrated facial landmark localization and super-resolution of real-world low resolution faces in arbitrary poses with gans,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 109–117

    2018

  20. [28]

    Deep laplacian pyramid networks for fast and accurate super-resolution,

    W.-S. Lai, J.-B. Huang, N. Ahuja, and M.-H. Yang, “Deep laplacian pyramid networks for fast and accurate super-resolution,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 624–632

    2017

  21. [29]

    Perceptual losses for real-time style transfer and super-resolution,

    J. Johnson, A. Alahi, and L. Fei-Fei, “Perceptual losses for real-time style transfer and super-resolution,” inComputer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11- 14, 2016, Proceedings, Part II 14. Springer, 2016, pp. 694–711

    2016

  22. [30]

    A stochastic approximation method,

    H. Robbins and S. Monro, “A stochastic approximation method,”The annals of mathematical statistics, pp. 400–407, 1951

    1951

  23. [31]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  24. [32]

    Enhanced deep residual networks for single image super-resolution,

    B. Lim, S. Son, H. Kim, S. Nah, and K. Mu Lee, “Enhanced deep residual networks for single image super-resolution,” inProceedings of the IEEE conference on computer vision and pattern recognition workshops, 2017, pp. 136–144

    2017

  25. [33]

    Decoupled weight decay regularization,

    I. Loshchilov, “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  26. [34]

    Adan: Adaptive nesterov momentum algorithm for faster optimizing deep models,

    X. Xie, P. Zhou, H. Li, Z. Lin, and S. Yan, “Adan: Adaptive nesterov momentum algorithm for faster optimizing deep models,”IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2024

    2024

  27. [35]

    Image super-resolution with an enhanced group convolutional neural network,

    C. Tian, Y . Yuan, S. Zhang, C.-W. Lin, W. Zuo, and D. Zhang, “Image super-resolution with an enhanced group convolutional neural network,” Neural Networks, vol. 153, pp. 373–385, 2022

    2022

  28. [36]

    Sharpened cosine distance as an alternative for convolutions,

    R. Pisoni, “Sharpened cosine distance as an alternative for convolutions,” 2022, https://rpisoni.dev/posts/cossim-convolution/

    2022

  29. [37]

    Ntire 2017 challenge on single image super-resolution: Dataset and study,

    E. Agustsson and R. Timofte, “Ntire 2017 challenge on single image super-resolution: Dataset and study,” inProceedings of the IEEE con- ference on computer vision and pattern recognition workshops, 2017, pp. 126–135

    2017

  30. [38]

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

    M. Bevilacqua, A. Roumy, C. Guillemot, and M. L. Alberi-Morel, “Low- complexity single-image super-resolution based on nonnegative neighbor embedding.” BMV A press, 2012

    2012

  31. [39]

    On single image scale-up using sparse-representations,

    R. Zeyde, M. Elad, and M. Protter, “On single image scale-up using sparse-representations,” inCurves and Surfaces: 7th International Con- ference, Avignon, France, June 24-30, 2010, Revised Selected Papers 7. Springer, 2012, pp. 711–730

    2010

  32. [40]

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

    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,” inProceedings eighth IEEE international conference on computer vision. ICCV 2001, vol. 2. ...

    2001

  33. [41]

    Single image super-resolution from transformed self-exemplars,

    J.-B. Huang, A. Singh, and N. Ahuja, “Single image super-resolution from transformed self-exemplars,” inProceedings of the IEEE confer- ence on computer vision and pattern recognition, 2015, pp. 5197–5206

    2015

  34. [42]

    Multi-scale attention network for single image super-resolution,

    Y . Wang, Y . Li, G. Wang, and X. Liu, “Multi-scale attention network for single image super-resolution,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 5950–5960

    2024

  35. [43]

    Residual dense net- work for image super-resolution,

    Y . Zhang, Y . Tian, Y . Kong, B. Zhong, and Y . Fu, “Residual dense net- work for image super-resolution,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2472–2481

    2018

  36. [44]

    Separable and reversible data hiding in encrypted images using parametric binary tree labeling,

    S. Yi and Y . Zhou, “Separable and reversible data hiding in encrypted images using parametric binary tree labeling,”IEEE Transactions on Multimedia, vol. 21, no. 1, pp. 51–64, 2018

    2018

  37. [45]

    A heterogeneous group cnn for image super-resolution,

    C. Tian, Y . Zhang, W. Zuo, C.-W. Lin, D. Zhang, and Y . Yuan, “A heterogeneous group cnn for image super-resolution,”IEEE transactions on neural networks and learning systems, 2022

    2022

  38. [46]

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

    N. Ahn, B. Kang, and K.-A. Sohn, “Fast, accurate, and lightweight super-resolution with cascading residual network,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 252–268

    2018

  39. [47]

    Fast and accurate single image super- resolution via information distillation network,

    Z. Hui, X. Wang, and X. Gao, “Fast and accurate single image super- resolution via information distillation network,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 723–731

    2018

  40. [48]

    A+: Adjusted anchored neighborhood regression for fast super-resolution,

    R. Timofte, V . De Smet, and L. Van Gool, “A+: Adjusted anchored neighborhood regression for fast super-resolution,” inComputer Vision– ACCV 2014: 12th Asian Conference on Computer Vision, Singapore, Singapore, November 1-5, 2014, Revised Selected Papers, Part IV 12. Springer,...

    2014

  41. [49]

    Jointly optimized regressors for image super-resolution,

    D. Dai, R. Timofte, and L. Van Gool, “Jointly optimized regressors for image super-resolution,” inComputer Graphics F orum, vol. 34, no. 2. Wiley Online Library, 2015, pp. 95–104

    2015

  42. [50]

    Fast and accurate image upscaling with super-resolution forests,

    S. Schulter, C. Leistner, and H. Bischof, “Fast and accurate image upscaling with super-resolution forests,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3791– 3799

    2015

  43. [51]

    Deep networks for image super-resolution with sparse prior,

    Z. Wang, D. Liu, J. Yang, W. Han, and T. Huang, “Deep networks for image super-resolution with sparse prior,” inProceedings of the IEEE international conference on computer vision, 2015, pp. 370–378

    2015

  44. [52]

    Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connec- tions,

    X. Mao, C. Shen, and Y .-B. Yang, “Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connec- tions,”Advances in neural information processing systems, vol. 29, 2016

    2016

  45. [53]

    Single image super-resolution with non-local means and steering kernel regression,

    K. Zhang, X. Gao, D. Tao, and X. Li, “Single image super-resolution with non-local means and steering kernel regression,”IEEE Transactions on Image Processing, vol. 21, no. 11, pp. 4544–4556, 2012

    2012

  46. [54]

    Trainable nonlinear reaction diffusion: A flexible framework for fast and effective image restoration,

    Y . Chen and T. Pock, “Trainable nonlinear reaction diffusion: A flexible framework for fast and effective image restoration,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1256– 1272, 2016

    2016

  47. [55]

    Fast single image super-resolution via dilated residual networks,

    Z. Lu, Z. Yu, P. Yali, L. Shigang, W. Xiaojun, L. Gang, and R. Yuan, “Fast single image super-resolution via dilated residual networks,”Ieee Access, vol. 7, pp. 109 729–109 738, 2018

    2018

  48. [56]

    Structure-preserving image super-resolution via contextualized multitask learning,

    Y . Shi, K. Wang, C. Chen, L. Xu, and L. Lin, “Structure-preserving image super-resolution via contextualized multitask learning,”IEEE transactions on multimedia, vol. 19, no. 12, pp. 2804–2815, 2017

    2017

  49. [57]

    Image super resolution based on fusing multiple convolution neural networks,

    H. Ren, M. El-Khamy, and J. Lee, “Image super resolution based on fusing multiple convolution neural networks,” inProceedings of the IEEE conference on computer vision and pattern recognition workshops, 2017, pp. 54–61

    2017

  50. [58]

    Beyond deep residual learning for image restoration: Persistent homology-guided manifold simplification,

    W. Bae, J. Yoo, and J. Chul Ye, “Beyond deep residual learning for image restoration: Persistent homology-guided manifold simplification,” inProceedings of the IEEE conference on computer vision and pattern recognition workshops, 2017, pp. 145–153

    2017

  51. [59]

    Self-learning super- resolution using convolutional principal component analysis and random matching,

    J. Xu, M. Li, J. Fan, X. Zhao, and Z. Chang, “Self-learning super- resolution using convolutional principal component analysis and random matching,”IEEE Transactions on Multimedia, vol. 21, no. 5, pp. 1108– 1121, 2018

    2018

  52. [60]

    New architecture of deep recursive convolution networks for super-resolution,

    F. Cao and B. Chen, “New architecture of deep recursive convolution networks for super-resolution,”Knowledge-Based Systems, vol. 178, pp. 98–110, 2019

    2019

  53. [61]

    Lightweight image super-resolution with enhanced cnn,

    C. Tian, R. Zhuge, Z. Wu, Y . Xu, W. Zuo, C. Chen, and C.-W. Lin, “Lightweight image super-resolution with enhanced cnn,”Knowledge- Based Systems, vol. 205, p. 106235, 2020

    2020

  54. [62]

    A dual cnn for image super-resolution,

    J. Song, J. Xiao, C. Tian, Y . Hu, L. You, and S. Zhang, “A dual cnn for image super-resolution,”Electronics, vol. 11, no. 5, p. 757, 2022

    2022

  55. [63]

    Unfolding the alternating optimization for blind super resolution,

    Y . Huang, S. Li, L. Wang, T. Tanet al., “Unfolding the alternating optimization for blind super resolution,”Advances in Neural Information Processing Systems, vol. 33, pp. 5632–5643, 2020

    2020

  56. [64]

    Deep constrained least squares for blind image super-resolution,

    Z. Luo, H. Huang, L. Yu, Y . Li, H. Fan, and S. Liu, “Deep constrained least squares for blind image super-resolution,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 17 642–17 652

    2022

  57. [65]

    Structure-aware deep networks and pixel- level generative adversarial training for single image super-resolution,

    W. Shi, F. Tao, and Y . Wen, “Structure-aware deep networks and pixel- level generative adversarial training for single image super-resolution,” IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1–14, 2023

    2023

  58. [66]

    Image super-resolution via lightweight attention-directed feature aggregation network,

    L. Wang, K. Li, J. Tang, and Y . Liang, “Image super-resolution via lightweight attention-directed feature aggregation network,”ACM Trans- actions on Multimedia Computing, Communications and Applications, vol. 19, no. 2, pp. 1–23, 2023

    2023

  59. [67]

    Acdmsr: Accelerated conditional diffusion models for single image super-resolution,

    A. Niu, T. X. Pham, K. Zhang, J. Sun, Y . Zhu, Q. Yan, I. S. Kweon, and Y . Zhang, “Acdmsr: Accelerated conditional diffusion models for single image super-resolution,”IEEE Transactions on Broadcasting, 2024

    2024

  60. [68]

    Cross-resolution feature attention network for image super-resolution,

    A. Liu, S. Li, and Y . Chang, “Cross-resolution feature attention network for image super-resolution,”The Visual Computer, vol. 39, no. 9, pp. 3837–3849, 2023

    2023

  61. [69]

    Sam- diffsr: Structure-modulated diffusion model for image super-resolution,

    C. Wang, Z. Hao, Y . Tang, J. Guo, Y . Yang, K. Han, and Y . Wang, “Sam- diffsr: Structure-modulated diffusion model for image super-resolution,” arXiv preprint arXiv:2402.17133, 2024

    2024 arXiv

  62. [70]

    Image quality metrics: Psnr vs. ssim,

    A. Hore and D. Ziou, “Image quality metrics: Psnr vs. ssim,” in2010 20th international conference on pattern recognition. IEEE, 2010, pp. 2366–2369

    2010

  63. [71]

    Image quality assessment: from error visibility to structural similarity,

    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, vol. 13, no. 4, pp. 600–612, 2004

    2004

  64. [72]

    Deep Learning Face Attributes in the Wild,

    Z. Liu, P. Luo, X. Wang, and X. Tang, “Deep Learning Face Attributes in the Wild,” inProceedings of the IEEE International Conference on Computer Vision, 2015, pp. 3730–3738

    2015

  65. [73]

    The PASCAL Visual Object Classes Challenge: A Retrospective,

    M. Everingham, S. M. A. Eslami, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “The PASCAL Visual Object Classes Challenge: A Retrospective,”International Journal of Computer Vision, vol. 111, no. 1, pp. 98–136, 2015

    2015

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.