pith. sign in

arxiv: 2605.08173 · v1 · submitted 2026-05-05 · 💻 cs.CV · cs.LG

CASISR: Circular Arbitrary-Scale Image Super-Resolution

Honggui Li , Zhengyang Zhang , Dingtai Li , Sinan Chen , Nahid Md Lokman Hossain , Xinfeng Xu , Yinlu Qin , Ruobing Wang
show 6 more authors
Hantao Lu Yuting Feng Maria Trocan Dimitri Galayko Amara Amara Mohamad Sawan
This is my paper

Pith reviewed 2026-05-12 01:15 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords arbitrary-scale image super-resolutionclosed-loop architecturegeneralization performanceimage reconstructiondegradation modelCASISR
0
0 comments X

The pith

The CASISR closed-loop architecture improves arbitrary-scale image super-resolution quality by combining the super-resolution step with its degradation reverse in a stable nonlinear feedback system.

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

The paper seeks to boost generalization in deep models for super-resolving images at any scale factor by closing the usual one-way pipeline into a circle that re-uses test data. Typical open-loop methods map low-resolution input directly to high-resolution output without checking consistency against how the output would degrade back down. By inserting a degradation model and forming a loop, the method lets reconstruction errors feed back to refine the result on unseen images. This matters for real images whose exact scale and degradation are unknown in advance, because the loop can correct mismatches without retraining. The authors supply a mathematical loop equation, prove it is reasonable with conditional probabilities and stable with Taylor approximations, and show concrete gains on fractional scales and high-edge images such as text.

Core claim

CASISR creates a closed-loop system in which an arbitrary-scale super-resolution model and a degradation model (bicubic, noisy, or learnable) operate together according to a nonlinear loop equation. Reasonableness follows from conditional probability theory; stability follows from Taylor-series approximation around the fixed point. First-order and second-order absolute difference images serve as new comparison metrics. Comprehensive experiments establish that the resulting reconstructions exceed those of eight prior state-of-the-art open-loop ASISR methods, with the largest gains appearing at fractional scale factors and on text or stripe patterns that contain abrupt edges.

What carries the argument

The circular closed-loop architecture that couples an arbitrary-scale super-resolution model to its degradation reverse inside a single nonlinear feedback equation.

If this is right

  • Any existing arbitrary-scale super-resolution network can be wrapped inside the CASISR loop to gain the reported quality lift without retraining.
  • Fractional scale factors benefit more than integer ones because the loop can iteratively correct the mismatch between training and test distributions.
  • Images containing sharp, high-frequency edges such as printed text or stripes show the clearest improvement in the first- and second-order difference metrics.
  • The loop equation supplies a concrete stability certificate that applies across the three common degradation types (bicubic, noisy, learnable).

Where Pith is reading between the lines

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

  • The same loop construction could be tried on other image-restoration tasks such as deblurring or denoising whenever a forward degradation model is available.
  • Because the loop refines each test image individually, it offers a form of test-time adaptation that may reduce reliance on ever-larger training collections.
  • If the iterations inside the loop converge in a few steps, the approach could be adapted for on-device or real-time super-resolution pipelines.

Load-bearing premise

The degradation model correctly reverses the super-resolution mapping at any scale without injecting new instability or artifacts when the two models run together in a loop.

What would settle it

Running the closed loop on a standard benchmark set produces reconstructions that are no better, or visibly worse, than the corresponding open-loop baseline on the same test images.

Figures

Figures reproduced from arXiv: 2605.08173 by Amara Amara, Dimitri Galayko, Dingtai Li, Hantao Lu, Honggui Li, Maria Trocan, Mohamad Sawan, Nahid Md Lokman Hossain, Ruobing Wang, Sinan Chen, Xinfeng Xu, Yinlu Qin, Yuting Feng, Zhengyang Zhang.

Figure 6
Figure 6. Figure 6: ΔPSNRa and ΔPSNRm of the proposed 8 closed-loop CASISR algorithms [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
read the original abstract

The generalization performance (GP) of deep learning-based arbitrary-scale image super-resolution (ASISR) methods is subject to limited training datasets and unlimited testing datasets. It is vitally significant to enhance the GP of the pretrained ASISR models by making full use of the testing samples. The ASISR models usually employ an open-loop architecture from low-resolution (LR) images to super-resolution (SR) images. The degradation model from SR samples to LR samples is known bicubic down-sampling for the classical ASISR, is supposed down-sampling with additive random noise for the blind ASISR, and is learnable for the real-world ASISR. Combining the ASISR and degradation models, it is potentially possible to adopt a closed-loop architecture based on the automatic control theory for strengthening the GP of the ASISR methods. Therefore, this paper proposes a closed-loop architecture, circular ASISR (CASISR), to lift the capability of image reconstruction. A mathematical nonlinear loop equation is established to describe the CASISR, the reasonability of the CASISR is proven by conditional probability theory, and the stability of the CASISR is proven by Taylor series approximation. The first-order and second-order absolute difference images are defined to compare the image reconstruction performance of the ASISR and the CASISR methods. Comprehensive simulation experiments show that the proposed CASISR approach outperforms the eight state-of-the-art ASISR approaches in the quality of image reconstruction. Especially, the proposed CASISR is extraordinarily suitable for fractional SR scale factors and is extremely effective for text and stripe images with drastically changed edges.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes CASISR, a closed-loop architecture for arbitrary-scale image super-resolution that combines pretrained ASISR models with known or learnable degradation models. It defines a nonlinear loop equation, proves reasonability via conditional probability theory, establishes stability via Taylor series approximation, and reports simulation results showing outperformance over eight SOTA ASISR methods, with particular gains for fractional scales and text/stripe images with sharp edges.

Significance. If the closed-loop construction proves stable and artifact-free for arbitrary scales, the approach would offer a general, training-free way to boost generalization of existing ASISR models by exploiting test samples, addressing a practical limitation in the field. The explicit use of control-theoretic ideas and the emphasis on fractional scales are potentially valuable contributions, though the local character of the stability argument restricts immediate impact.

major comments (3)
  1. [stability proof] Stability proof (Taylor series section): The analysis linearizes the nonlinear loop equation locally around an operating point. For non-integer fractional scales the ASISR+degradation composition is strongly nonlinear and the degradation operator can easily map outside the linearization neighborhood; no global convergence guarantee or bound on deviation is provided, undermining the claim that the closed loop remains stable for arbitrary scales.
  2. [simulation experiments] Experiments and evaluation: The reported outperformance on text and stripe images is presented without quantitative error analysis, ablation isolating the closed-loop contribution, or verification that high-frequency artifacts are not introduced at fractional scales. This is load-bearing for the central claim that CASISR is “extraordinarily suitable” for such cases.
  3. [degradation model] Degradation model assumption: The construction assumes the degradation operator exactly inverts the forward process (bicubic, noisy, or learnable). In the noisy and real-world regimes explicitly tested, this inversion is only approximate; the paper does not quantify how mismatch propagates through the loop or affects the conditional-probability argument.
minor comments (2)
  1. [evaluation metrics] The first- and second-order absolute difference images are introduced for comparison but their precise definitions and relation to standard metrics (PSNR/SSIM/LPIPS) are not stated explicitly.
  2. [mathematical formulation] Notation for the nonlinear loop equation could be clarified by explicitly listing all operators and variables in a single displayed equation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments, which help clarify the scope and limitations of our work. We address each major comment below, indicating where revisions will be made to the manuscript.

read point-by-point responses

  1. Referee: Stability proof (Taylor series section): The analysis linearizes the nonlinear loop equation locally around an operating point. For non-integer fractional scales the ASISR+degradation composition is strongly nonlinear and the degradation operator can easily map outside the linearization neighborhood; no global convergence guarantee or bound on deviation is provided, undermining the claim that the closed loop remains stable for arbitrary scales.

    Authors: We agree that the Taylor-series stability argument is local, as is standard for analyzing nonlinear systems via linearization in control theory. The manuscript establishes local asymptotic stability around the operating point defined by the fixed point of the loop equation. While we do not claim a global convergence proof, the extensive simulations across integer and non-integer scales (including fractional factors) demonstrate practical convergence without divergence or visible artifacts. We will revise the stability section to explicitly state the local character of the result, add a brief discussion of the conditions under which the linearization remains valid, and include a note on the absence of a global guarantee together with empirical evidence of robustness. revision: partial

  2. Referee: Experiments and evaluation: The reported outperformance on text and stripe images is presented without quantitative error analysis, ablation isolating the closed-loop contribution, or verification that high-frequency artifacts are not introduced at fractional scales. This is load-bearing for the central claim that CASISR is “extraordinarily suitable” for such cases.

    Authors: The current manuscript reports PSNR/SSIM gains on standard benchmarks and provides visual comparisons for text/stripe images. We acknowledge that an explicit ablation isolating the closed-loop feedback and quantitative artifact analysis (e.g., high-frequency error maps or edge sharpness metrics) at fractional scales would strengthen the claims. In the revision we will add (i) an ablation study comparing open-loop ASISR versus CASISR on the same backbone, (ii) quantitative error analysis including first- and second-order difference statistics for the text/stripe subset, and (iii) verification that no spurious high-frequency artifacts are introduced at non-integer scales. revision: yes

  3. Referee: Degradation model assumption: The construction assumes the degradation operator exactly inverts the forward process (bicubic, noisy, or learnable). In the noisy and real-world regimes explicitly tested, this inversion is only approximate; the paper does not quantify how mismatch propagates through the loop or affects the conditional-probability argument.

    Authors: The conditional-probability justification in the paper is derived under the modeling assumption that the degradation operator is the exact inverse of the forward imaging process. For the classical bicubic case this holds exactly; for noisy and real-world cases we employ the best available approximate models. We will add a new subsection that quantifies sensitivity to model mismatch via a first-order perturbation analysis around the nominal degradation operator and will include additional experiments that vary noise levels to illustrate how residual mismatch affects loop convergence and reconstruction quality. revision: yes

Circularity Check

2 steps flagged

CASISR loop equation and stability proof are defined directly from ASISR+degradation composition with local Taylor approximation

specific steps
  1. self definitional [Abstract]
    "Combining the ASISR and degradation models, it is potentially possible to adopt a closed-loop architecture based on the automatic control theory for strengthening the GP of the ASISR methods. Therefore, this paper proposes a closed-loop architecture, circular ASISR (CASISR), to lift the capability of image reconstruction. A mathematical nonlinear loop equation is established to describe the CASISR, the reasonability of the CASISR is proven by conditional probability theory, and the stability of the CASISR is proven by Taylor series approximation."

    CASISR and its loop equation are defined by direct composition of the input ASISR model and degradation model; the 'proof' of reasonability via conditional probability and the performance lift are therefore constructed from the same components rather than derived from independent first principles.

  2. other [Abstract]
    "the stability of the CASISR is proven by Taylor series approximation"

    Stability is established only by local linearization (Taylor series) of the nonlinear loop equation around a presumed fixed point; this does not independently guarantee global convergence or artifact-free behavior for arbitrary fractional scales where the composition is highly nonlinear.

full rationale

The paper constructs the CASISR architecture and its governing nonlinear loop equation explicitly by combining the existing ASISR model with the degradation model (bicubic/noisy/learnable). Reasonability is asserted via conditional probability on that same composition, and stability is shown only via first-order Taylor linearization around an operating point. While experiments provide independent empirical comparison, the central claims of improved GP and suitability for fractional scales rest on this self-composed loop and its local approximation rather than an independent derivation or global guarantee. This creates moderate circularity burden without fully reducing the result to tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on combining known ASISR models with degradation models into a stable closed loop; no new free parameters or invented entities are explicitly introduced beyond the loop equation itself.

axioms (2)
  • domain assumption Degradation from SR to LR is accurately captured by bicubic, noisy, or learnable models
    Invoked to close the feedback loop and enable the nonlinear equation.
  • ad hoc to paper The closed-loop system remains stable under the stated approximations
    Proven via Taylor series but depends on the specific loop gain and model accuracy.

pith-pipeline@v0.9.0 · 5634 in / 1249 out tokens · 50833 ms · 2026-05-12T01:15:39.187096+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages

  1. [1]

    Liu Hongying, Li Zekun, Shang Fanhua, Liu Yuanyuan, Wan Liang, Feng Wei, Timofte Radu, Arbitrary-scale Super-resolution via Deep Learning: A Comprehensive Survey , INFORMA TION FUSION, V olume 102, DOI10.1016/j.inffus.2023.102015, Article Number 102015, 2024

    work page arXiv 2023

  2. [2]

    Huang Yihao, Luo Xin, Guo Qing, Juefei-Xu Felix, Jia Xiaojun, Miao Weikai, Pu Geguang, Liu Yang, Scale-Invariant Adversarial Attack Against Arbitrary -Scale Super -Resolution, IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, V olume 20, Page 3909- 3924, DOI10.1109/TIFS.2025.3550079, 2025

    work page arXiv 2025

  3. [3]

    P ARK HYUNJUN, Improvement of Mask-RCNN Performance Using Deep -Learning-Based Arbitrary-Scale Super -Resolution Module , JOURNAL of the K OREA INSTITUTE OF INFORMATION AND COMMUNICATION ENGINEERING, V olume 26, Issue 3, Page 381-388, DOI10.6109/jkiice.2022.26.3.381, 2022

    work page 2022

  4. [4]

    Pavlovic Milena, Al Hajj Ghadi S., Kanduri Chakravarthi, Pensar Johan, Wood Mollie E., Sollid Ludvig M., Greiff Victor, Sandve Geir K., Improving generalization of machine learning-identified biomarkers using causal modelling with examples from immune receptor diagnostics, NATURE MACHINE INTELLIGENCE, V olume 6, Issue 1, Page 15-24, DOI10.1038/s42256-023-...

    work page 2024

  5. [5]

    Liu Liyuan, Chen Yaohui, Li Weifu, Wang Yingjie, Gu Bin, Zheng Feng, Chen Hong, Generalization bounds of deep neural networks with τ-mixing samples, IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, V olume 36, Issue 8, Page 14596-14610 DOI10.1109/TNNLS.2025.3526235, 2025

    work page arXiv 2025

  6. [6]

    Li Juncheng, Pei Zehua, Li Wenjie, Gao Guangwei, Wang Longguang, Wang Yingqian, Zeng Tieyong, A Systematic Survey of Deep Learning -Based Single -Image Super -Resolution, ACM COMPUTING SURVEYS, V olume 56, Issue 10, DOI10.1145/3659100, 2024

    work page 2024

  7. [7]

    Galimberti Clara Lucia, Furieri Luca, Ferrari-Trecate Giancarlo, Parametrizations of all stable closed-loop responses: From theory to neural network control design , ANNUAL REVIEWS IN CONTROL, V olume 60, DOI10.1016/j.arcontrol.2025.101012, Article Number 101012, 2025

    work page arXiv 2025

  8. [8]

    Li Honggui, Hossain Nahid Md Lokman, Trocan Maria, Galayko Dimitri, Sawan Mohamad, CMISR: Circular medical image super -resolution, ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, V olume 133, Part B, DOI10.1016/j.engappai.2024.108222, Article Number 108222, 2024

    work page arXiv 2024

  9. [9]

    Honggui Li, Zhengyang Zhang, Dingtai Li, Sinan Chen, Nahid Md Lokman Hossain, Xinfeng Xu, Yuting Feng, Hantao Lu, Yinlu Qin, Ruobing Wang, Maria Trocan, Dimitri Galayko, Amara, Mohamad Sawan, CECGSR: Circular ECG Super -Resolution, arXiv preprint, arXiv:2508.11658, Page 1-42, 2025

    work page arXiv 2025

  10. [10]

    Li Honggui, Li Dingtai, Chen Sinan, Hossain Nahid Md Lokman, Xu Xinfeng, Feng Yuting, Lu Hantao, Trocan Maria, Galayko Dimitri, Amara Amara, Sawan Mohamad, CCSMRI: Circular compressed sensing MRI, BIOMEDICAL SIGNAL PROCESSING AND CONTROL, V olume 111, Article Number 108355, Page 1-15, DOI10.1016/j.bspc.2025.108355, 2026

    work page arXiv 2025

  11. [11]

    Li Honggui, Chen Sinan, Hossain Nahid Md Lokman, Trocan Maria, Galayko Dimitri, Sawan Mohamad, CIC: Circular image compression, arXiv preprint, arXiv:2407.15870, Page 1-33, 2024

    work page arXiv 2024

  12. [12]

    Wan Wenbo, Wang Zezhu, Wang Zhiyan, Gu Lingchen, Sun Jiande, Wang Qiang, Arbitrary- Scale Image Super -Resolution via Degradation Perception , IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING, V olume10, Page 666-676, DOI10.1109/TCI.2024.3393712, 2024

    work page arXiv 2024

  13. [13]

    Chen Honggang, He Xiaohai, Yang Hong, Wu Yuanyuan, Qing Linbo, Sheriff Ray E., Self- 41 supervised cycle-consistent learning for scale -arbitrary real-world single image super -resolution, EXPERT SYSTEMS WITH APPLICATIONS , V olume 212, DOI10.1016/j.eswa.2022.118657, Article Number 118657, 2023

    work page arXiv 2022

  14. [14]

    Yan Jin, Chen Zongren, Pei Zhiyuan, Lu Xiaoping, Zheng Hua, MambaSR: Arbitrary-Scale Super-Resolution Integrating Mamba with Fast Fourier Convolution Blocks , MATHEMATICS, V olume 12, Issue 15, DOI10.3390/math12152370, Article Number 2370, 2024

    work page 2024

  15. [15]

    Fang Jing, Yu Yinbo, Wang Zhongyuan, Ding Xin , Hu Ruimin, An Image Arbitrary -Scale Super-Resolution Network Using Frequency -domain Information, ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS, V olume 20, Issue 3, DOI10.1145/3616376, Article Number 81, 2024

    work page 2024

  16. [16]

    Akita Kazutoshi, Ukita Norimichi, Test-Time Cost-and-Quality Controllable Arbitrary-Scale Super-Resolution with Variable Fourier Components , IEEE OPEN JOURNAL OF SIGNAL PROCESSING, V olume 6, Page1017-1030, DOI10.1109/OJSP.2025.3602742, 2025

    work page arXiv 2025

  17. [17]

    Wang Shibo, Xing Yuming, Shi Shengzhu, Guo Zhichang, A Taylor expansion-based texture and edge-preserving interpolation approach for arbitrary -scale image super-resolution, PATTERN RECOGNITION, V olume 169, DOI10.1016/j.patcog.2025.111965, Article Number 111965, 2025

    work page arXiv 2025

  18. [18]

    Pang Kaifeng, Zhao Kai, Hung Alex Ling Yu, Zheng Haoxin, Y an Ran, Sung Kyunghyun, NExpR: Neural Explicit Representation for fast arbitrary -scale medical image super -resolution, COMPUTERS IN BIOLOGY AND MEDICINE, V olume 184, Page 109354, DOI10.1016/j.compbiomed.2024.109354, 2025

    work page arXiv 2024

  19. [19]

    Luo Laigan, Yi Benshun, Wang Zhongyuan, He Zheng, Zhu Chao, Bidirectional scale-aware upsampling network for arbitrary -scale video super -resolution, IMAGE AND VISION COMPUTING, V olume 148, DOI10.1016/j.imavis.2024.105116, Article Number 105116, 2024

    work page arXiv 2024

  20. [20]

    Lyu Jun , Wang Guangming , Wang Zi , Dong Shunjie , Ding Weiping , Wang Chengyan , Diffusion-prior based implicit neural representation for arbitrary -scale cardiac cine MRI super - resolution, INFORMA TION FUSION, V olume 126, Part A, DOI10.1016/j.inffus.2025.103510, Article Number 103510, 2025

    work page arXiv 2025

  21. [21]

    Wei Jinbao, Y ang Gang, Wei Wei, Liu Aiping, Chen Xun, Multi-Contrast MRI Arbitrary-Scale Super-Resolution via Dynamic Implicit Network , IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY , V olume 35, Issue 9, Page 8973-8988, DOI10.1109/TCSVT.2025.3556210, 2025

    work page arXiv 2025

  22. [22]

    Tu Xinyue, Li Guangyuan, Liu Yuanxing, Synergizing frequency domain and texture -aware implicit module for MRI arbitrary-scale super-resolution, BIOMEDICAL SIGNAL PROCESSING AND CONTROL, V olume 98, DOI10.1016/j.bspc.2024.106760, Article Number 106760, 2024

    work page arXiv 2024

  23. [23]

    Wu Qing, Li Yuwei, Sun Y awen, Zhou Yan, Wei Hongjiang, Yu Jingyi, Zhang Yuyao, An Arbitrary Scale Super-Resolution Approach for 3D MR Images via Implicit Neural Representation, IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, V olume 27, Issue 2, Page 1004-1015, DOI10.1109/JBHI.2022.3223106, 2023

    work page arXiv 2022

  24. [24]

    Tsai Yi-Ting, Implicit Image Function Modeling for Arbitrary-Resolution and Arbitrary-Scale Face Super-Resolution, National Yang Ming Chiao Tung University, Dissertation/Thesis, 2023

    work page 2023

  25. [25]

    Wen Caizhen, Yang Zhijing, Shi Yukai, Qing Chunmei, Cheng Yongqiang, Enhanced implicit function-based network for arbitrary-scale image super-resolution, JOURNAL OF ELECTRONIC IMAGING, V olume 31, Issue 4, DOI10.1117/1.JEI.31.4.043015, 2022

    work page 2022

  26. [26]

    Wang Shuangxing, Wang Lihui, Cao Ying, Deng Zeyu, Ye Chen, Wang Rongpin, Zhu Yuemin, 42 Wei Hongjiang , Self-supervised arbitrary -scale super -angular resolution diffusion MRI reconstruction, MEDICAL PHYSICS, V olume 52, Issue 5, Page 2976-2998, DOI10.1002/mp.17691, 2025

    work page 2025

  27. [27]

    Chen Chuan, Wang Weiwei, Jia Xixi, Feng Xiangchu, Wei Hanjia, Local Gaussian ensemble for arbitrary -scale image super -resolution, COMPUTER VISION AND IMAGE UNDERSTANDING, V olume 257, DOI10.1016/j.cviu.2025.104372, Article Number 104372, 2025

    work page arXiv 2025

  28. [28]

    Chen Guochao , Nie Jiangtao , Wei Wei , Zhang Lei , Zhang Yanning , Arbitrary-Scale Hyperspectral Image Super -Resolution from a Fusion Perspective with Spatial Priors , IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING , V olume 62, DOI10.1109/TGRS.2024.3481041, Article Number 5536611, 2024

    work page arXiv 2024

  29. [29]

    Maciel Corbin, Zou Qing, FDoSR-Net: Frequency-Domain Informed Auto-Encoder Network for Arbitrary-Scale 3D Whole-Heart MRI Super-Resolution, BIOENGINEERING-BASEL, V olume 12, Issue 2, DOI10.3390/bioengineering12020129, Article Number 129, 2025

    work page 2025

  30. [30]

    Rong Yi, Jia Mingbin, Zhan Yufei, Zhou Luoyu, SR-RDFAN-LOG: Arbitrary-scale logging image super-resolution reconstruction based on residual dense feature aggregation, GEOENERGY SCIENCE AND ENGINEERING, V olume 240, DOI10.1016/j.geoen.2024.213042, Article Number 213042, 2024

    work page arXiv 2024

  31. [31]

    Gurrola-Ramos Javier, Alarcon Teresa E., Dalmau Oscar, Arbitrary Scale Super -Resolution Neural Network Based on Residual Channel -Spatial Attention, IEEE ACCESS, V olume 10, Page 108697-108709, DOI10.1109/ACCESS.2022.3211302, 2022

    work page arXiv 2022

  32. [32]

    Fu Ying, Chen Jian, Zhang Tao, Lin Yonggang, Residual scale attention network for arbitrary scale image super -resolution, NEUROCOMPUTING, V olume 427, Page 201-211, DOI10.1016/j.neucom.2020.11.010, 2021

    work page 2020

  33. [33]

    Shen Jialiang , Wang Yucheng, Zhang Jian , ASDN: A Deep Convolutional Network for Arbitrary Scale Image Super-Resolution, MOBILE NETWORKS & APPLICATIONS, V olume 26, Issue 1, Page 13-26, Special Issue, DOI10.1007/s11036-020-01720-2, 2021

    work page 2021

  34. [34]

    Li Guangping, Xiao Huanling , Liang Dingkai, Ling Bingo Wing -Kuen, Multi-scale cross- fusion for arbitrary scale image super resolution, MULTIMEDIA TOOLS AND APPLICATIONS, DOI10.1007/s11042-024-18677-z, Early Access, 2024

    work page 2024

  35. [35]

    Zhao Yaoqian, Teng Qizhi, Chen Honggang, Zhang Shujiang, He Xiaohai, Li Yi, Sheriff Ray E., Activating More Information in Arbitrary -Scale Image Super -Resolution, IEEE TRANSACTIONS ON MULTIMEDIA , V olume 26, Page 7946-7961, DOI10.1109/TMM.2024.3373257, 2024

    work page arXiv 2024

  36. [36]

    He Zhi, He Dan, A Unified Network for Arbitrary Scale Super -Resolution of Video Satellite Images, IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, V olume 59, Issue 10, Page 8812-8825, DOI10.1109/TGRS.2020.3038653, 2021

    work page arXiv 2020

  37. [37]

    Zhu Jin, Tan Chuan, Yang Junwei, Yang Guang, Lio Pietro, Arbitrary Scale Super-Resolution for Medical Images, INTERNATIONAL JOURNAL OF NEURAL SYSTEMS, V olume 31, Issue 10, DOI10.1142/S0129065721500374, Article Number 2150037, 2021

    work page 2021

  38. [38]

    Zhu Jinchen, Zhang Mingjian, Zheng Ling, Weng Shizhuang, Multi-scale implicit transformer with re-parameterization for arbitrary-scale super-resolution, PATTERN RECOGNITION, V olume 162, DOI10.1016/j.patcog.2024.111327, Article Number 111327, 2025

    work page arXiv 2024

  39. [39]

    Jiang Shuguo , Li Nanying , Xu Meng , Zhang Shuyu , Jia Sen , SQformer: Spectral -Query Transformer for Hyperspectral Image Arbitrary-Scale Super-Resolution, IEEE TRANSACTIONS 43 ON GEOSCIENCE AND REMOTE SENSING , V olume 62, DOI10.1109/TGRS.2024.3463745, Article Number 5532315, 2024

    work page arXiv 2024

  40. [40]

    Kim Min Hyuk, Yoo Seok Bong, Memory-Efficient Discrete Cosine Transform Domain Weight Modulation Transformer for Arbitrary-Scale Super-Resolution, MATHEMATICS, V olume 11, Issue 18, DOI10.3390/math11183954, Article Number 3954, 2023

    work page 2023

  41. [41]

    Zhao Ming, Fang Jia, Chen Boyang, An Arbitrary-Scale Super-Resolution Network for Multi- Contrast MRI With Permuted Cross -Attention, IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, V olume 29, Issue 8, Page 5762-5772, DOI10.1109/JBHI.2025.3548696, 2025

    work page arXiv 2025

  42. [42]

    Liu Lanqing, Zou Jing, Xu Cheng, Wang Kang, Lyu Jun, Xu Xuemiao, Hu Zhanli, Qin Jing, IM-Diff: Implicit Multi-Contrast Diffusion Model for Arbitrary Scale MRI Super-Resolution, IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, V olume 29, Issue 6, Page 4200- 4212, DOI10.1109/JBHI.2025.3544265, 2025

    work page arXiv 2025

  43. [43]

    Han Zhitao, Huang Wenhui, Arbitrary scale super-resolution diffusion model for brain MRI images, COMPUTERS IN BIOLOGY AND MEDICINE , V olume 170, DOI10.1016/j.compbiomed.2024.108003, Article Number 108003, 2024

    work page arXiv 2024

  44. [44]

    Hwang Inje, Lee Woojin, Kim Munchurl, LIIFusion: Learning Implicit Image Function Using Image Prior Generated by an Efficient Diffusion Model for Arbitrary-Scale Image Super-Resolution, JOURNAL OF BROADCAST ENGINEERING , V olume 29, Issue 6, Page 981-998, DOI10.5909/JBE.2024.29.6.981, 2024

    work page 2024

  45. [45]

    Cui Jizhou , Dong Wenqian, Qu Jiahui , Wu Xiaoyang, Xiao Song , Li Yunsong, CFMDM: Coarse-to-Fine Meta-Diffusion Model for Scale -Arbitrary Hyperspectral Super-Resolution, IEEE GEOSCIENCE AND REMOTE SENSING LETTERS , V olume 21, DOI10.1109/LGRS.2024.3369026, Article Number 5503705, 2024

    work page arXiv 2024

  46. [46]

    Wang Yufeng, Zhang Huayu , Zeng Xianlin , Wang Bowen, Li Wei, Ding Wenrui, Binary Lightweight Neural Networks for Arbitrary Scale Super -Resolution of Remote Sensing Images , IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING , V olume 63, DOI10.1109/TGRS.2025.3529696, Article Number 5609716, 2025

    work page arXiv 2025

  47. [47]

    Jia Haoran, Zhao Pengjie, Cao Tongtai, Wang Xin, Liu Yue, Lightweight Arbitrary-Scale Super- Resolution via Texture-Aware deformation, OPTICS AND LASER TECHNOLOGY, V olume 192, Part E, DOI10.1016/j.optlastec.2025.113922, Article Number 113922, 2025

    work page arXiv 2025

  48. [48]

    He Gang , Wu Chang , Quan Guancheng , Lai Xinquan , Li Yunsong , SA-CVSR: Scale Compressed Video Super -Resolution, PATTERN RECOGNITION , V olume 167, DOI10.1016/j.patcog.2025.111745, Article Number 111745, 2025

    work page arXiv 2025

  49. [49]

    Duan Minghong, Qu Linhao, Yang Zhiwei, Wang Manning, Zhang Chenxi, Song Zhijian, An efficient dual -branch framework via implicit self -texture enhancement for arbitrary -scale histopathology image super -resolution, SCIENTIFIC REPORTS , V olume 15, Issue 1, DOI10.1038/s41598-025-02503-z, Article Number 18883, 2025

    work page 2025

  50. [50]

    Li Guangping, Xiao Huanling, Liang Dingkai, Enhanced dual branches network for arbitrary- scale image super -resolution, ELECTRONICS LETTERS , V olume 59, Issue 1, DOI10.1049/ell2.12689, 2023

    work page 2023

  51. [51]

    Zhang Menglei, Ling Qiang, Bilateral Upsampling Network for Single Image Super-Resolution with Arbitrary Scaling Factors, IEEE TRANSACTIONS ON IMAGE PROCESSING, V olume 30, Page 4395-4408, DOI10.1109/TIP.2021.3071708, 2021. 44

    work page arXiv 2021

  52. [52]

    Fang Jing, Xiao Jing, Wang Xu, Chen Dan, Hu Ruimin, Arbitrary Scale Super Resolution Network for Satellite Imagery , CHINA COMMUNICATIONS, V olume19, Issue8, Page234-246, 2022

    work page 2022

  53. [53]

    Truong Anh Minh, Philips Wilfried, Veelaert Peter, Depth Completion and Super-Resolution with Arbitrary Scale Factors for Indoor Scenes , SENSORS, V olume 21, Issue 14, DOI10.3390/s21144892, Article Number 4892, 2021

    work page 2021

  54. [54]

    An Tai, Huo Chunlei, Xiang Shiming, Pan Chunhong, Combining Discrete and Continuous Representation: Scale -Arbitrary Super -Resolution for Satellite Images , REMOTE SENSING , V olume 15, Issue 7, DOI10.3390/rs15071827, Article Number 1827, 2023

    work page 2023

  55. [55]

    Li Zhiheng, Li Muheng, Fan Jixuan, Chen Lei, Tang Yansong, Lu Jiwen, Zhou Jie, Learning Dual-Level Deformable Implicit Representation for Real-World Scale Arbitrary Super-Resolution, Lecture Notes in Computer Science (LNCS), V olume 15127, Page 352-368, 2025, European Conference on Computer Vision (ECCV), 2024

    work page 2025

  56. [56]

    Guancheng Zhou, Yisi Luo, Xile Zhao, Deyu Meng, Efficient Arbitrary-Scale Image Super- Resolution via Functional Tensor Decomposition , IEEE TRANSACTIONS ON MULTIMEDIA, Early Access, 2026

    work page 2026

  57. [57]

    Oh Jong Kwon, Son Hwijae, Hwang Hyung Ju, Oh Jihyong, SoFoNO: Arbitrary-scale image super-resolution via Sobolev Fourier neural operator, Neurocomputing, V olume 661, 131944, 2026

    work page 2026

  58. [58]

    Chen Du, Chen Liyi, Zhang Zhengqiang, Zhang Lei, Generalized and Efficient 2D Gaussian Splatting for Arbitrary -scale Super -Resolution, International Conference on Computer Vision (ICCV), 2025

    work page 2025

  59. [59]

    Alexander Becker, Rodrigo Caye Daudt, Dominik Narnhofer, Torben Peters, Nando Metzger, Jan Dirk Wegner, Konrad Schindler , Thera: Aliasing-Free Arbitrary-Scale Super-Resolution with Neural Heat Fields, TRANSACTIONS ON MACHINE LEARNING RESEARCH (TMLR), 1-24, 2025

    work page 2025

  60. [60]

    Fu Huiyuan, Peng Fei, Li Xianwei, Li Yejun, Wang Xin, Ma Huadong, Continuous Optical Zooming: A Benchmark for Arbitrary -Scale Image Super -Resolution in Real World , IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    work page 2024

  61. [61]

    Shang Wei , Zhang Wanying , Gu Shuhang , Zhu Pengfei , Hu Qinghua , Ren Dongwei , BasicA VSR: Arbitrary-Scale Video Super -Resolution via Image Priors and Enhanced Motion Compensation, arXiv preprint arXiv:2510.26149, 2025

    work page arXiv 2025

  62. [62]

    Terre des hommes

    Longguang Wang, Yingqian Wang, Zaiping Lin, Jungang Yang, Wei An, and Yulan Guo , Learning A Single Network for Scale -Arbitrary Super -Resolution, International Conference on Computer Vision (ICCV), 2021. Statements and Declarations 1 Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscri...

    work page 2021