Pith. sign in

REVIEW 3 major objections 6 minor 55 references

Test-time Cost-and-Quality Controllable Arbitrary-Scale Super-Resolution with Variable Fourier Components

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a single arbitrary-scale super-resolution network in which a recurrent neural network sequentially outputs Fourier components, so that choosing how many recurrences to run at test time trades computational cost against…

desk verdict A clever RNN-based Fourier component mechanism gives real test-time quality scaling, but the cost-control half of the claim needs actual FLOPs/latency numbers and less handicapped baselines before it holds up. read the letter →

arxiv 2412.05517 v2 pith:SVBCQLB3 submitted 2024-12-07 eess.IV cs.CV

Kazutoshi Akita , Norimichi Ukita This is my paper · ORCID
classification eess.IVcs.CV
keywords arbitrary-scalesuper-resolutioncost-qualitycontrolrecurrentneuralnetworkFouriercomponentstest-timevariable-lengthtrainingsingle-imagePSNR
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

This paper tries to establish that arbitrary-scale super-resolution can be made cost-and-quality controllable in a single trained model by casting reconstruction as a sum of Fourier components that a recurrent network emits one at a time. The key claim is that the number of RNN recurrences $T$ can be chosen at inference: fewer components cost less and degrade gracefully, more components improve PSNR, and the same weights serve every budget. This matters because previous arbitrary-scale SR methods either fix the number of Fourier components in a fully connected layer or require retraining a new model for each complexity level, which prevents end users from trading quality for compute after deployment. Experiments on DIV2K, BSD100, and Urban100 report that PSNR rises monotonically with $T$ and that at reduced $T$ the method outperforms state-of-the-art baselines modified for variable component counts.

What carries the argument

The central object is the recurrent Fourier predictor, an implicit function that starts from a latent code $z$ and local offset $\delta$, runs an RNN cell for $T$ steps, and at each step estimates an amplitude vector $A_t\in\mathbb{R}^2$, a frequency matrix $F_t\in\mathbb{R}^2$, and phase information $h_p(c,A_t,F_t)$. The additive identity that the pixel value is the sum of the individual Fourier contributions is the mechanism that allows variable-length output with fixed weights, since no fully connected decoder with fixed input width is involved. The second machine is random-length training, which samples $T\sim U(1,T_{\max})$ each iteration and forces the early recurrences to carry the most dominant Fourier information, making truncation graceful rather than catastrophic.

What would settle it

Run the released model at $T=4,8,16,32,64$ on the same hardware and measure end-to-end latency and FLOPs; if total compute is nearly flat across $T$, or PSNR stops increasing with $T$ on held-out datasets, the claimed cost-and-quality controllability fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the additive nature of Fourier components lets a recurrent predictor turn super-resolution into a controllable, progressive reconstruction. Instead of decoding a fixed set of Fourier components through a final layer, an RNN estimates one amplitude--frequency pair (with phase) per recurrence, and the SR pixel value is simply the sum over recurrences, written as $I^{HR}(x_q)=\sum_{t=1}^{T} A_t \cdot [\cos(\pi F_t \delta + h_p(c,A_t,F_t)),\ \sin(\pi F_t \delta + h_p(c,A_t,F_t))]$. Because the number of recurrences $T$ is chosen at test time, one network spans a quality--cost curve. The paper also establishes that this only works if training draws $T$ uniformly from $1$ to $T_{\max}$; training only at $T_{\max}$ makes test-time truncation unstable. In the reported comparison, at $T=64$ the LTE baseline has a slightly higher PSNR, but across $T=4,\dots,32$ the proposed method keeps substantially higher PSNR than the modified baselines, with PSNR growing monotonically as $T$ increases.

Load-bearing premise

The load-bearing premise is that the number of RNN recurrences $T$ is a faithful proxy for computational cost: the paper reports no FLOPs, latency, or memory measurements, so if the fixed feature extractor or per-step overhead dominates inference, cutting $T$ may not save the user meaningful cost.

Editorial extensions

If this is right

  • A single trained SR model can cover multiple deployment budgets: an end user picks $T$ at inference and gets a monotonically ordered PSNR without downloading or switching to another network.
  • The training procedure suggests a general recipe for recurrent generative models whose outputs are additive: random-length training makes truncation tolerable, so the model can be used at any prefix length.
  • Because test quality peaks near the training maximum $T_{\max}$, the deployment target can be chosen by training with the largest budget that will actually be used, and then any $T\le T_{\max}$ remains available.
  • The comparison implies that fixed fully connected Fourier decoders are not safely truncatable, so a recurrent additive decoder is a design choice rather than an optional refinement.
  • Relative position encoding materially improves low-$T$ performance, indicating that relations learned among early recurrences transfer to later recurrences and help sparse reconstructions.

Reading between the lines

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

  • The paper's cost claim rests on the unmeasured assumption that total inference cost scales with $T$; a direct FLOP and latency measurement across $T$ values would turn the stated quality--cost trade-off into a quantitative curve, which the paper leaves open.
  • The recurrence order likely encodes a learned coarse-to-fine decomposition, so probing which spatial frequencies are added at each step could test whether early recurrences carry low-frequency content and motivate progressive SR for streaming.
  • Because the reconstruction is a pure sum, one could decode a shared prefix of components at multiple scales or continue the recurrence to refine an existing SR image, both of which the paper does not explore.
  • Random-length training may transfer to other additive output representations such as wavelets or Laplacian pyramids, an implication the paper only hints at through its Fourier choice.
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 a test-time cost-and-quality (CQ) controllable arbitrary-scale super-resolution method. It uses an RNN to sequentially estimate Fourier components (amplitude and frequency), and reconstructs the SR image by summing the contributions of the selected number of components T (Eq. 11). The RNN is trained with a variable number of recurrences (Eq. 12) to generalize across different T at test time. Experiments on DIV2K, BSD100, and Urban100 show that PSNR degrades gracefully as T is reduced, whereas truncated fixed-length baselines (LTE, CiaoSR, CLIT) drop sharply. The paper claims to be the first to enable test-time CQ control in arbitrary-scale SR without retraining.

Significance. The idea of using an RNN for variable-length Fourier-component prediction is novel and directly addresses a real deployment need: adapting SR quality and compute to varying resource constraints. The training strategy with uniformly sampled T is simple and appears effective; Figure 8 provides a clear ablation showing the benefit of random-length training. If the cost-saving claim is substantiated, this would be a practical contribution to edge and streaming SR applications. However, the paper currently lacks any measurement of computational cost, and its comparison to modified baselines is not a fair test of CQ controllability. The quality-side results are solid and reproducible in principle, but the central cost-control claim remains unverified.

major comments (3)
  1. [Section IV (Results)] The central claim of test-time cost-and-quality control is only validated on the quality axis. The abstract, introduction, and Figure 1 explicitly state that fewer recurrences lead to lower computational cost, but Section IV reports no FLOPs, latency, memory, or energy measurements. The total inference cost includes the EDSR feature extractor, feature unfolding, and bilinear upsampling, which are independent of T; if these dominate, the actual savings from reducing T may be small. Please provide a measured cost-quality trade-off curve (e.g., FLOPs or wall-clock time per image vs PSNR) for T in {1..64}, and explicitly report the fixed-cost share. Without this, the cost-control half of the contribution is unsupported.
  2. [Section IV-B (Comparison with SoTA methods)] The baseline comparison is not a fair evaluation of CQ controllability. LTE, CiaoSR, and CLIT are trained with a fixed number of Fourier components (K=64) and only at test time have their last FC layer removed and a subset of components selected. Because they are never trained with a variable-length objective like Eq. (12), the large PSNR drops at T<64 are an expected consequence of train-test mismatch. To support the claim of "lower PSNR drop than other state-of-the-art methods," the baselines should either be retrained with the same variable-length training strategy (preserving their architectures) or the comparison should be explicitly framed as "naive truncation of fixed-length models" rather than as a CQ-controllable baseline comparison.
  3. [Section IV-B (Random selection)] The "Random" selection procedure for the modified baselines is not a cost-control method. It averages the output over 100 random subsets, which multiplies the inference cost by 100, so it does not correspond to any meaningful operating point on a cost-quality curve. The paper should report only the deterministic "Descending" selection for the baseline comparisons, or provide cost measurements for all variants so that the cost-quality trade-off can be compared fairly.
minor comments (6)
  1. [Equation (14)] Equation (14) has mismatched parentheses in the cosine and sine arguments; the arguments should read cos(πFtδ + hp(c, At, Ft)) and sin(πFtδ + hp(c, At, Ft)).
  2. [Section III-B] The text describes F_t as a "frequency matrix" but Eq. (11) treats it as a vector in R^2; please clarify the dimension and use a dot product notation such as πF_t · δ.
  3. [Section IV-B] The dataset name "Urban109" appears to be a typo for "Urban100", matching the reference [51].
  4. [Section V (Conclusion)] The sentence "we plan to explore autoregressive models [] instead of RNNs" contains an empty citation; please fill it in or remove the bracket.
  5. [Figure 8] In Figure 8, the x-axis extends to 128 Fourier components, but Tmax is 64; clarify whether a model trained with Tmax=64 is being evaluated beyond its training range (which would degrade) or whether an additional model is used.
  6. [Section IV-C (Position Encoding)] The abbreviation "SPE" is used without definition; please spell out "Sinusoidal Positional Encoding" or whatever the intended expansion is when it first appears.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reconstruction is an explicit sum of Fourier components, PSNR behavior is measured empirically, and the cost-side gap is missing validation rather than circular reasoning.

full rationale

I walked the derivation chain from Eq. (11), where the SR image is defined as an explicit sum of T Fourier components estimated by RNN recurrences, through the variable-length training objective in Eq. (12), which samples T uniformly during training. The reported monotonic improvement of PSNR with T and the comparison against modified baselines are empirical measurements, not quantities fitted and then renamed as predictions. No equation or fitted parameter forces the PSNR results; the model could in principle fail to generalize across T, and the paper tests that generalization. The "cost" half of the claim is asserted via recurrence count rather than measured with FLOPs, latency, or memory, but that is a missing-validation and correctness concern, not a circular reduction: no definition, fitting step, or self-citation makes the cost claim true by construction. Self-citations such as DBPN [2] appear only as background related work and are not load-bearing for the proposed method or its evaluation. Accordingly, the paper is self-contained with respect to its quality claims, and no circular step is exhibited.

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

The central claim depends on hand-chosen hyperparameters (Tmax, RNN depth, position encoding, scale range) and on the fairness of the modified SoTA baselines. No free parameters are fitted to a theoretical target, and no new physical or mathematical entities are posited.

free parameters (5)
  • Tmax (maximum recurrences) = 64
    Set equal to LTE's K=64; defines the upper bound of the CQ range and is swept in Table 6. The Tmax=4 model behaves very differently, so Tmax materially affects the claimed monotonic quality behavior.
  • RNN depth and type = 4-layer Linear Transformer
    Chosen by hand and not ablated; affects both quality and per-recurrence cost, which matters for the cost-control claim.
  • Position encoding = SPE relative encoding
    Table 7 shows relative encoding improves low-T PSNR; this is a post-experiment design choice baked into the final model.
  • Training scale range = Uniform in [1,4]
    All tested scales (1.4 to 4.0) lie in this range; the 'arbitrary-scale' claim is not demonstrated beyond 4, so the range is a hand-chosen scope limit.
  • Training epochs and LR schedule = 400 epochs, LR 1e-4 halved every 200
    Standard but unablated; contributes to final quality and is part of the implicit recipe for reproducing the reported PSNR.
assumptions (5)
  • standard math A 2D signal can be reconstructed by summing sinusoids with estimated frequency, amplitude, and phase (Eq. 14).
    Fourier representation; standard mathematical background and not questioned.
  • domain assumption Bicubic downsampling of HR images produces realistic LR inputs, and DIV2K training generalizes to BSD100 and Urban100.
    Standard SR benchmark practice; the paper evaluates on these datasets without testing real-world LR degradation.
  • ad hoc to paper An RNN trained with random recurrence counts T~U(1,Tmax) generalizes to any T in that range at test time.
    This is the core training-strategy assumption; Figure 8 and Table 6 support it empirically for the Tmax=64 model, but it is not guaranteed in general and fails for extrapolation beyond Tmax.
  • ad hoc to paper Removing the last FC layer of LTE/CiaoSR/CLIT and selecting a subset of Fourier components is a fair way to make them CQ-controllable for comparison.
    This comparison design is introduced in Section IV-B; it handicaps methods trained with fixed K and is not a standard baseline for variable-length Fourier SR.
  • domain assumption A pixel-wise reconstruction loss (L1 or L2) is used for training, though not stated in the paper.
    Training details in Section IV-A omit the loss function; the reported quality depends on it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Test-time Cost-and-Quality Controllable Arbitrary-Scale Super-Resolution with Variable Fourier Components." pith.science (2026). https://pith.science/paper/SVBCQLB3

@misc{pith2026241205517,
  author       = {Pith},
  title        = {Pith review of: Test-time Cost-and-Quality Controllable Arbitrary-Scale Super-Resolution with Variable Fourier Components},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SVBCQLB3}},
  note         = {Machine review of arXiv:2412.05517}
}
read the original abstract

Super-resolution (SR) with arbitrary scale factor and cost-and-quality controllability at test time is essential for various applications. While several arbitrary-scale SR methods have been proposed, these methods require us to modify the model structure and retrain it to control the computational cost and SR quality. To address this limitation, we propose a novel SR method using a Recurrent Neural Network (RNN) with the Fourier representation. In our method, the RNN sequentially estimates Fourier components, each consisting of frequency and amplitude, and aggregates these components to reconstruct an SR image. Since the RNN can adjust the number of recurrences at test time, we can control the computational cost and SR quality in a single model: fewer recurrences (i.e., fewer Fourier components) lead to lower cost but lower quality, while more recurrences (i.e., more Fourier components) lead to better quality but more cost. Experimental results prove that more Fourier components improve the PSNR score. Furthermore, even with fewer Fourier components, our method achieves a lower PSNR drop than other state-of-the-art arbitrary-scale SR methods.

Figures

Figures reproduced from arXiv: 2412.05517 by the authors.

Figure 1
Figure 1. FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FIGURE 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: FIGURE 5 [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: FIGURE 6 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIGURE 7 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 50 canonical work pages

  1. [1]

    Enhanced deep resid- ual networks for single image super-resolution,

    B. Lim, S. Son, H. Kim, S. Nah, and K. M. Lee, “Enhanced deep resid- ual networks for single image super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 1132–1140, June 2017

    work page 2017

  2. [2]

    Deep back-projection networks for super-resolution,

    M. Haris, G. Shakhnarovich, and N. Ukita, “Deep back-projection networks for super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 1664– 1673, June 2018

    work page 2018

  3. [3]

    Wavelet domain style transfer for an effective perception-distortion tradeoff in single im- age super-resolution,

    X. Deng, R. Yang, M. Xu, and P. L. Dragotti, “Wavelet domain style transfer for an effective perception-distortion tradeoff in single im- age super-resolution,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 3076–3085, June 2019

    work page 2019

  4. [4]

    Srflow: Learning the super-resolution space with normalizing flow,

    A. Lugmayr, M. Danelljan, L. V . Gool, and R. Timofte, “Srflow: Learning the super-resolution space with normalizing flow,” in Pro- ceedings of the European Conference on Computer Vision , pp. 715– 732, Springer, 2020

    work page 2020

  5. [5]

    PAMS: quantized super-resolution via parameterized max scale,

    H. Li, C. Yan, S. Lin, X. Zheng, B. Zhang, F. Yang, and R. Ji, “PAMS: quantized super-resolution via parameterized max scale,” in Proceedings of the European Conference on Computer Vision , 2020

    work page 2020

  6. [6]

    Learning with privileged information for efficient image super-resolution,

    W. Lee, J. Lee, D. Kim, and B. Ham, “Learning with privileged information for efficient image super-resolution,” in Proceedings of the European Conference on Computer Vision , 2020

    work page 2020

  7. [7]

    Latticenet: Towards lightweight image super-resolution with lattice block,

    X. Luo, Y . Xie, Y . Zhang, Y . Qu, C. Li, and Y . Fu, “Latticenet: Towards lightweight image super-resolution with lattice block,” in Proceedings of the European Conference on Computer Vision , 2020

    work page 2020

  8. [8]

    Image super-resolution by neural texture transfer,

    Z. Zhang, Z. Wang, Z. L. Lin, and H. Qi, “Image super-resolution by neural texture transfer,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , June 2019

    work page 2019

Show all 55 references
  1. [9]

    Structure- preserving super resolution with gradient guidance,

    C. Ma, Y . Rao, Y . Cheng, C. Chen, J. Lu, and J. Zhou, “Structure- preserving super resolution with gradient guidance,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, June 2020

    2020

  2. [10]

    Meta-sr: A magnification-arbitrary network for super-resolution,

    X. Hu, H. Mu, X. Zhang, Z. Wang, T. Tan, and J. Sun, “Meta-sr: A magnification-arbitrary network for super-resolution,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1575–1584, June 2019

    2019

  3. [11]

    Learning A single network for scale-arbitrary super-resolution,

    L. Wang, Y . Wang, Z. Lin, J. Yang, W. An, and Y . Guo, “Learning A single network for scale-arbitrary super-resolution,” in Proceed- ings of the IEEE/CVF International Conference on Computer Vision , pp. 4781–4790, October 2021

    2021

  4. [12]

    CUF: continuous upsampling fil- ters,

    C. N. Vasconcelos, A. C. ¨Oztireli, M. J. Matthews, M. Hashemi, K. Swersky, and A. Tagliasacchi, “CUF: continuous upsampling fil- ters,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9999–10008, June 2023

    2023

  5. [13]

    Deep arbitrary-scale image super-resolution via scale-equivariance pursuit,

    X. Wang, X. Chen, B. Ni, H. Wang, Z. Tong, and Y . Liu, “Deep arbitrary-scale image super-resolution via scale-equivariance pursuit,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1786–1795, June 2023

    2023

  6. [14]

    Learning continuous image repre- sentation with local implicit image function,

    Y . Chen, S. Liu, and X. Wang, “Learning continuous image repre- sentation with local implicit image function,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 8628–8638, June 2021

    2021

  7. [15]

    Local texture estimator for implicit repre- sentation function,

    J. Lee and K. H. Jin, “Local texture estimator for implicit repre- sentation function,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1919–1928, June 2022

    1919

  8. [16]

    Neural fields with thermal activations for arbitrary-scale super- resolution,

    A. Becker, R. C. Daudt, N. Metzger, J. D. Wegner, and K. Schindler, “Neural fields with thermal activations for arbitrary-scale super- resolution,” arXiv:abs/2311.17643, 2023

    2023

  9. [17]

    Ciaosr: Continuous implicit attention- in-attention network for arbitrary-scale image super-resolution,

    J. Cao, Q. Wang, Y . Xian, Y . Li, B. Ni, Z. Pi, K. Zhang, Y . Zhang, R. Timofte, and L. V . Gool, “Ciaosr: Continuous implicit attention- in-attention network for arbitrary-scale image super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern...

    2023

  10. [18]

    Cascaded local implicit transformer for arbitrary-scale super-resolution,

    H. Chen, Y . Xu, M. Hong, Y . Tsai, H. Kuo, and C. Lee, “Cascaded local implicit transformer for arbitrary-scale super-resolution,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18257–18267, June 2023

    2023

  11. [19]

    Ultrasr: Spatial encoding is a missing key for implicit image function-based arbitrary-scale super-resolution,

    X. Xu, Z. Wang, and H. Shi, “Ultrasr: Spatial encoding is a missing key for implicit image function-based arbitrary-scale super-resolution,” arXiv:2103.12716, 2021

    2021 arXiv

  12. [20]

    Enhancing multi-scale implicit learn- ing in image super-resolution with integrated positional encoding,

    Y . Liu, Y . Guo, and S. Zhang, “Enhancing multi-scale implicit learn- ing in image super-resolution with integrated positional encoding,” arXiv:2112.05756, 2021

    2021 arXiv

  13. [21]

    Implicit neural representations with periodic activation functions,

    V . Sitzmann, J. N. P. Martel, A. W. Bergman, D. B. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation functions,” in Proceedings of the Advances in Neural Information Processing Systems, December 2020

    2020

  14. [22]

    Neural fourier filter bank,

    Z. Wu, Y . Jin, and K. M. Yi, “Neural fourier filter bank,” in Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14153–14163, June 2023

    2023

  15. [23]

    Bacon: Band- limited coordinate networks for multiscale scene representation,

    D. B. Lindell, D. V . Veen, J. J. Park, and G. Wetzstein, “Bacon: Band- limited coordinate networks for multiscale scene representation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 16231–16241, June 2022

    2022

  16. [24]

    Improvement of JPEG XL lossy image coding using region adaptive DCT block partitioning structure,

    J. Cho, O. Kwon, and S. Choi, “Improvement of JPEG XL lossy image coding using region adaptive DCT block partitioning structure,” IEEE Access, vol. 9, pp. 113213–113225, 2021

    2021

  17. [25]

    Compression of hyperspectral image using pca–dct technology,

    R. J. Yadav and M. S. Nagmode, “Compression of hyperspectral image using pca–dct technology,” in Innovations in Electronics and Communication Engineering (H. S. Saini, R. K. Singh, and K. S. Reddy, eds.), pp. 269–277, 2018

    2018

  18. [26]

    Intriguing findings of frequency selection for image deblurring,

    X. Mao, Y . Liu, F. Liu, Q. Li, W. Shen, and Y . Wang, “Intriguing findings of frequency selection for image deblurring,” in The pro- ceedings of Association for the Advancement of Artifical intelligence (B. Williams, Y . Chen, and J. Neville, eds.), pp. 1905–1913, February 2023

    1905

  19. [27]

    Spatial-frequency mutual learning for face super-resolution,

    C. Wang, J. Jiang, Z. Zhong, and X. Liu, “Spatial-frequency mutual learning for face super-resolution,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 22356– 22366, June 2023

    2023

  20. [28]

    Fourier space losses for efficient perceptual image super-resolution,

    D. Fuoli, L. V . Gool, and R. Timofte, “Fourier space losses for efficient perceptual image super-resolution,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 2340–2349, October 2021

    2021

  21. [29]

    Srwarp: Generalized image super-resolution under arbitrary transformation,

    S. Son and K. M. Lee, “Srwarp: Generalized image super-resolution under arbitrary transformation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 7782– 7791, June 2021

    2021

  22. [30]

    Multi-scale fusion for arbitrary- scale image super-resolution,

    H. Weng, J. Lin, G. Lin, and Y . Niu, “Multi-scale fusion for arbitrary- scale image super-resolution,” in The Proceedings of the International Conference on Electronic Information Technology and Computer En- gineering, October 2019. 10 VOLUME , <Society logo(s) and publicatio...

    2019

  23. [31]

    On the spectral bias of neural networks,

    N. Rahaman, A. Baratin, D. Arpit, F. Draxler, M. Lin, F. A. Hamprecht, Y . Bengio, and A. C. Courville, “On the spectral bias of neural networks,” in Proceedings of the International Conference on Machine Learning, vol. 97, pp. 5301–5310, June 2019

    2019

  24. [32]

    Deep mutual learning,

    Y . Zhang, T. Xiang, T. M. Hospedales, and H. Lu, “Deep mutual learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 4320–4328, June 2018

    2018

  25. [33]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” in Proceedings of the Advances in Neural Information Processing Systems, December 2015

    2015

  26. [34]

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,

    S. Han, H. Mao, and W. J. Dally, “Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding,” in Proceedings of the International Conference on Learning Representations, May 2016

    2016

  27. [35]

    Learning both weights and connections for efficient neural network.,

    S. Han, J. Pool, J. Tran, and W. J. Dally, “Learning both weights and connections for efficient neural network.,” in Proceedings of the Advances in Neural Information Processing Systems , December 2015

    2015

  28. [36]

    Xnor-net: Imagenet classification using binary convolutional neural networks.,

    M. Rastegari, V . Ordonez, J. Redmon, and A. Farhadi, “Xnor-net: Imagenet classification using binary convolutional neural networks.,” in Proceedings of the European Conference on Computer Vision , October 2016

    2016

  29. [37]

    Binaryconnect: Training deep neural networks with binary weights during propagations.,

    M. Courbariaux, Y . Bengio, and J.-P. David, “Binaryconnect: Training deep neural networks with binary weights during propagations.,” in Proceedings of the Advances in Neural Information Processing Systems, December 2015

    2015

  30. [38]

    Snip: single-shot network pruning based on connection sensitivity,

    N. Lee, T. Ajanthan, and P. H. S. Torr, “Snip: single-shot network pruning based on connection sensitivity,” in Proceedings of the Inter- national Conference on Learning Representations , May 2019

    2019

  31. [39]

    SOSP: efficiently capturing global correlations by second-order structured pruning,

    M. Nonnenmacher, T. Pfeil, I. Steinwart, and D. Reeb, “SOSP: efficiently capturing global correlations by second-order structured pruning,” in Proceedings of the International Conference on Learning Representations, OpenReview.net, April 2022

    2022

  32. [40]

    Group fisher pruning for practical network compression,

    L. Liu, S. Zhang, Z. Kuang, A. Zhou, J. Xue, X. Wang, Y . Chen, W. Yang, Q. Liao, and W. Zhang, “Group fisher pruning for practical network compression,” in Proceedings of the International Conference on Machine Learning (M. Meila and T. Zhang, eds.), vol. 139, pp. 7021–7032, ...

    2021

  33. [41]

    A survey on deep neural net- work pruning-taxonomy, comparison, analysis, and recommendations,

    H. Cheng, M. Zhang, and J. Q. Shi, “A survey on deep neural net- work pruning-taxonomy, comparison, analysis, and recommendations,” arXiv:2308.06767, 2023

    2023 arXiv

  34. [42]

    A comprehensive survey on model quantization for deep neural networks in image clas- sification,

    B. Rokh, A. Azarpeyvand, and A. Khanteymoori, “A comprehensive survey on model quantization for deep neural networks in image clas- sification,” ACM Transactions on Intelligent Systems and Technology , vol. 14, no. 6, pp. 97:1–97:50, 2023

    2023

  35. [43]

    Multi-scale dense networks for resource efficient image classification,

    G. Huang, D. Chen, T. Li, F. Wu, L. van der Maaten, and K. Q. Weinberger, “Multi-scale dense networks for resource efficient image classification,” in Proceedings of the International Conference on Learning Representations, OpenReview.net, May 2018

    2018

  36. [44]

    Fractalnet: Ultra- deep neural networks without residuals,

    G. Larsson, M. Maire, and G. Shakhnarovich, “Fractalnet: Ultra- deep neural networks without residuals,” in Proceedings of the Inter- national Conference on Learning Representations , OpenReview.net, April 2017

    2017

  37. [45]

    Learning anytime predictions in neural networks via adaptive loss balancing,

    H. Hu, D. Dey, M. Hebert, and J. A. Bagnell, “Learning anytime predictions in neural networks via adaptive loss balancing,” in The proceedings of Association for the Advancement of Artifical intelli- gence, pp. 3812–3821, February 2019

    2019

  38. [46]

    Memory-friendly scalable super-resolution via rewinding lottery ticket hypothesis,

    J. Lin, X. Luo, M. Hong, Y . Qu, Y . Xie, and Z. Wu, “Memory-friendly scalable super-resolution via rewinding lottery ticket hypothesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14398–14407, June 2023

    2023

  39. [47]

    Transformers are rnns: Fast autoregressive transformers with linear attention,

    A. Katharopoulos, A. Vyas, N. Pappas, and F. Fleuret, “Transformers are rnns: Fast autoregressive transformers with linear attention,” in Proceedings of the International Conference on Machine Learning , vol. 119, pp. 5156–5165, July 2020

    2020

  40. [48]

    Linearized relative positional encoding,

    Z. Qin, W. Sun, K. Lu, H. Deng, D. Li, X. Han, Y . Dai, L. Kong, and Y . Zhong, “Linearized relative positional encoding,” Transactions on Machine Learning Research , vol. 2023, 2023

    2023

  41. [49]

    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,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 1122– 1131, June 2017

    2017

  42. [50]

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

    D. R. Martin, C. C. Fowlkes, D. Tal, and J. Malik, “A database of human segmented natural images and its application to evaluat- ing segmentation algorithms and measuring ecological statistics,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 41...

    2001

  43. [51]

    Single image super-resolution from transformed self-exemplars,

    J. Huang, A. Singh, and N. Ahuja, “Single image super-resolution from transformed self-exemplars,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 5197– 5206, June 2015

    2015

  44. [52]

    A length-extrapolatable transformer,

    Y . Sun, L. Dong, B. Patra, S. Ma, S. Huang, A. Benhaim, V . Chaud- hary, X. Song, and F. Wei, “A length-extrapolatable transformer,” in Proceedings of the Annual Meeting of the Association for Compu- tational Linguistics (A. Rogers, J. L. Boyd-Graber, and N. Okazaki, eds.), p...

    2023

  45. [53]

    Length extrapolation of transformers: A survey from the perspective of position encoding,

    L. Zhao, X. Feng, X. Feng, B. Qin, and T. Liu, “Length extrapolation of transformers: A survey from the perspective of position encoding,” arXiv:2312.17044, 2023

    2023 arXiv

  46. [54]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proceedings of the Advances in Neural Information Processing Systems (I. Guyon, U. von Luxburg, S. Bengio, H. M. Wallach, R. Fergus, S. V . N. ...

    2017

  47. [55]

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

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” in Proceedings of the International Con...

    1999

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.