Accelerating Diffusion-based Super-Resolution with Dynamic Time-Spatial Sampling

Bin Wang; Chao Zhou; Haowei Zhu; Ming Sun; Qijie Wang; Rui Qin

REVIEW 5 major objections 6 minor 42 references

Accelerating Diffusion-based Super-Resolution with Dynamic Time-Spatial Sampling

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Diffusion super-resolution can run in half the steps by concentrating denoising iterations on the early and late timesteps where high-frequency detail is recovered.

desk verdict TDS is a solid empirical acceleration trick for diffusion SR, but the spatial component doesn't do the mechanistic work claimed and the paper's strongest claims outrun the evidence. read the letter →

arxiv 2505.12048 v2 pith:PE636MZM submitted 2025-05-17 cs.CV

Rui Qin , Qijie Wang , Ming Sun , Haowei Zhu , Chao Zhou , Bin Wang This is my paper · ORCID

classification cs.CV

keywords diffusionmodelssuper-resolutiontraining-freeaccelerationtimestepsamplingspatialadaptivehigh-frequencyrecoveryperceptualqualityinference

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that existing diffusion-based super-resolution methods waste iterations by sampling denoising steps uniformly, even though high-frequency image details are actually recovered in two concentrated windows: the earliest denoising steps and the final ones. On that basis it proposes Time-Spatial-aware Sampling (TSS), a training-free schedule that places more denoising steps in those critical windows and varies the schedule pixel by pixel according to local texture. The authors claim that attaching TSS to models such as SUPIR, PASD, and StableSR improves perceptual quality, with MUSIQ gains of 0.2 to 3.0, and matches or beats accelerated baselines such as ResShift and SinSR using about half the steps.

What carries the argument

The Time Dynamic Sampling schedule is built from a two-piece polynomial resampling function $$f(t,a,n,T,T') = \frac{t^n}{$a^{{n-1}}$}\; (t<a), \quad T - \frac{(T-t)^n}{(T-a)^{n-1}}\; (t\geq a),$$ which concentrates sampling density near $t=0$ and $t=T$ when $n>1$ and collapses to uniform sampling as $n\to 1$. Spatial Dynamic Sampling computes a smoothed local variance map $V_g$ from the low-resolution grayscale image, maps it linearly to per-pixel values of $a$ and $n$, and builds a per-pixel timestep tensor; the resulting spatial timestep embedding $t_{\text{emb}}^{\text{spatial}}$ is added to the network's main branch features at every spatial location, replacing the usual single global embedding. Together these two mechanisms let a pre-trained denoiser run one forward pass with a spatially varying timestep, allocating more steps to high-frequency recovery in textured regions while keeping smooth regions on a more uniform schedule.

What would settle it

Run TSS on a pre-trained diffusion super-resolution model while forcing large per-pixel timestep gaps between adjacent smooth and textured regions: if the spatially varying timestep embedding produces visible boundary artifacts or worse perceptual scores than uniform sampling, the central mechanism fails. Alternatively, measuring high-frequency SNR across a backbone outside the Stable Diffusion family and showing that the early/late two-stage pattern disappears would undercut the temporal motivation for TDS.

Watch

Extended reading notes

Core claim

The central discovery is a two-stage temporal pattern in the denoising of high-frequency signals: the signal-to-noise ratio for high-frequency components rises visibly in the early iterations (timestep roughly 1000 to 700) and again in the late iterations (roughly 400 to 0), while the middle iterations can actually add noise. Spatially, smooth regions finish denoising early, whereas textured regions, which carry more high-frequency content, concentrate their recovery in the final steps and benefit from a more non-uniform schedule. The paper claims that by exploiting this pattern, Time Dynamic Sampling redistributes steps toward the two critical windows, and Spatial Dynamic Sampling gives each pixel its own timestep schedule based on local variance, delivered through a spatially varying timestep embedding, so that pre-trained diffusion super-resolution models produce sharper textures with fewer total iterations and no additional training.

Load-bearing premise

The load-bearing assumption is that a pre-trained denoising network, trained with one global noise level per image, accepts and correctly interprets a spatially varying per-pixel timestep embedding, even when neighboring pixels are at very different noise levels.

Editorial extensions

If this is right

TSS attaches to pre-trained diffusion super-resolution models with no training, so any future improvement in the base model immediately inherits the speedup.
At seven steps, SUPIR with TSS beats the fifteen-step ResShift and the one-step SinSR on NIQE, MUSIQ, and QAlign on most benchmarks, showing a step reduction of more than half.
TSS combines with standard accelerated samplers such as DDIM and UniPC, improving all perceptual metrics over each sampler alone.
The non-uniform schedule can be trigonometric, exponential, or polynomial, indicating that the exact functional form is not essential as long as sampling concentrates on the early and late stages.

Reading between the lines

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

Beyond the paper: the same two-window sampling idea could accelerate other diffusion restoration tasks such as inpainting, deblurring, and text-to-image generation, which the authors name as future work.
Beyond the paper: since Spatial Dynamic Sampling keys off the local variance of the low-resolution image, replacing variance with a semantic or saliency map is a testable extension that may handle smooth-but-structured regions better.
Beyond the paper: the reported artifacts from large neighboring timestep gaps imply an upper bound on spatial timestep contrast for pre-trained networks, so training models with spatial timestep embeddings could lift that bound.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The short version: the time-step resampling part (TDS) is a real and useful observation. Concentrating reverse-process steps at the beginning and end consistently improves no-reference perceptual scores for StableSR, SUPIR, and PASD at the same step count, and the early-stage ablation in Tab. 3 supports the claim that early steps matter for high-frequency recovery. The spatial part (SDS) is the weak link. The per-pixel timestep embedding in Eq. 9 does not actually make the network denoise each pixel at a different noise level, because the latent z has one global noise level and the pre-trained network was trained on a single scalar embedding. The ablation data agree: SDS adds almost nothing over TDS in Tab. 2, and at 14 steps it makes NIQE slightly worse (3.59 to 3.65). The authors' own A.9 concedes boundary artifacts when neighboring timestep embeddings are far apart. So the mechanism described in Sec. 3.2 is closer to a content-adaptive modulation than per-pixel denoising, and the "spatial" novelty rests on a marginal effect.

What the paper does well: extensive evaluation across three anchor models and multiple real-world benchmarks, a direct comparison with DM-NonUniform, and an honest limitation appendix. The TDS result is robust across non-uniform functions and shows the expected behavior across models. The efficiency numbers (2.2x speedup for PASD at 7 steps) are concrete and reproducible.

Main soft spots besides SDS: no error bars anywhere, so I can't tell whether the small deltas are meaningful. Hyperparameters are tuned per model (Tab. 5), which is fitting to the test set even if not training. The abstract's "0.2-3.0 MUSIQ improvement" has exceptions: WebPhoto SUPIR drops from 60.42 to 58.72, and full-reference PSNR often drops (e.g., PASD on DIV2K from 21.28 to 20.98). This is a perceptual trade-off, not a free lunch.

Who this is for: people working on inference-time acceleration for diffusion SR. It deserves a serious referee, not a desk reject, but acceptance should hinge on releasing code, adding error bars, and rewriting the SDS claims to match what the module actually does.

Referee Report

5 major / 6 minor

Summary. The manuscript proposes Time-Spatial-aware Sampling (TSS), a training-free inference-time acceleration strategy for diffusion-based super-resolution. TSS combines Time Dynamic Sampling (TDS), a non-uniform polynomial timestep schedule that concentrates denoising steps in early and late iterations, with Spatial Dynamic Sampling (SDS), which modifies the timestep embedding spatially according to local variance. The method is motivated by an SNR analysis of SUPIR on RealPhoto60 and is evaluated by integrating TSS into StableSR, SUPIR, and PASD across six benchmarks with perceptual metrics and a full-reference appendix. The paper claims state-of-the-art perceptual quality with fewer steps than existing acceleration methods such as ResShift and SinSR.

Significance. If the claims hold, TSS would be a useful contribution because it is training-free, compatible with existing samplers, and shows consistent perceptual improvements in many of the reported settings. The paper has clear strengths: a comprehensible analysis of frequency- and spatial-domain denoising dynamics, extensive evaluation across three diffusion SR anchors, multiple datasets and metrics, additional FID results, comparison with DM-NonUniform, and a candid limitation discussion in A.9. The main weaknesses are an overstated universal MUSIQ claim, a spatial component (SDS) whose mechanism and measured contribution are not convincingly established, and per-model hyperparameter tuning that weakens the generality claim. These issues are addressable in revision, so the manuscript is not beyond repair.

major comments (5)

[§3.2, Eq. (9), A.9] The SDS mechanism is load-bearing for the claim that TSS is 'spatial' and training-free, but it is not convincingly established. At each reverse step the denoising network receives one latent z at a single global noise level, so the per-pixel embeddings Emb(t_i,j) do not describe the actual noise level of any pixel; the pretrained network was trained only with a global scalar timestep. The paper's own limitation statement in A.9 concedes that large neighboring timestep intervals can harm quality, which is consistent with SDS being an untrained content-adaptive perturbation rather than a true per-pixel denoiser. The empirical support in Table 2 is also weak: at Step=14, adding SDS to TDS makes NIQE worse (3.59 to 3.65) and changes MUSIQ by only +0.10, with no error bars. I request a control experiment that isolates whether the timestep semantics matter, such as replacing Emb(t_i,j) with a smooth variance-derived spatial bias of similar magnitude, and multiple-seed or paired results for the SDS ablation.
[Abstract, §4.2, Table 1] The abstract's claim of 'improving MUSIQ scores by 0.2–3.0' is not supported by the reported tables. In Table 1 (WebPhoto, SUPIR row), SUPIR-TSS MUSIQ is 58.72 versus 60.42 for SUPIR, a 1.70 drop; in DRealSR, PASD-TSS MUSIQ is 57.58 versus 58.21 for PASD. Several other rows show smaller gains within the noise level of the metric. The abstract, introduction, and conclusion should be qualified to state average or most-benchmark improvements, or the exceptions should be explicitly explained.
[A.2, Table 5, A.9] The hyperparameters of TDS and SDS are tuned per anchor model; A.2 states that 'small differences in the values of the hyperparameters are taken to achieve the best performance,' and A.9 acknowledges sensitivity to model architecture. Since the reported gains are the basis of the generality claim, the paper should report the performance of TSS with a single fixed hyperparameter set across all anchors and datasets, or provide a systematic sensitivity analysis for each anchor rather than only for PASD in Table 10.
[§4.3, Tables 1, 2, 11] No error bars or significance tests are reported, although some differences are small (e.g., Step=14 MUSIQ gain of +0.10 in Table 2, and several full-reference metrics that worsen in Table 11, such as PASD on RealSR PSNR dropping from 25.35 to 24.50). Because the central claim is that TSS 'consistently' improves metrics, the paper should provide variance across repeated runs or at least bootstrap confidence intervals for the key comparisons.
[§3.1, Table 1] The empirical motivation in Section 3.1 is based on SUPIR alone, yet TSS is applied to StableSR and PASD as well. The claim that the discovered temporal and spatial dynamics are generic to diffusion-based SR would be strengthened by repeating the frequency/spatial SNR analysis on at least one additional anchor model, especially since the hyperparameter ranges differ across models in Table 5.

minor comments (6)

[A.2, Table 5] The table header contains a typo: 'amin, nmax' should be 'amin, amax'; the rows for StableSR and SUPIR also use inconsistent formatting.
[Eq. (5)] Equation (5) contains 'n(vvgi,j)' with a duplicated 'v'; it should be 'n(vgi,j)'.
[Eq. (6)] The index ranges i in [0, H] and j in [0, W] should be [0, H-1] and [0, W-1] for a grid of size H x W.
[A.5, Table 8] The text in A.5 says 'As shown in Tab. 9' when referring to the DM-NonUniform comparison, but the table is labeled Table 8; all cross-references should be checked and corrected.
[Fig. 10] Several figure captions in the appendix contain formatting errors, including 'NIQE=,7.59' and 'NIQE=12.00 MUSIQ=26.57' with missing commas or spaces.
[§4.1, figure captions] The text repeatedly refers to 'the full version' for implementation details and additional results; if the full version is the arXiv appendix, the references should say 'Appendix' consistently.

Circularity Check

2 steps flagged · score 3.0 of 10

TSS's own HF-SNR analysis is self-confirming (Eq. 3 allocates samples to the stages it measures), and the headline gains are fitted via per-model hyperparameters on the reported benchmarks (Table 5); however, external no-reference metrics make the central SOTA claim non-tautological.

self definitional [Sec. 3.1-3.2 and Fig. 7 (Time Dynamic Sampling / HF SNR analysis)]
"high-frequency components uniquely exhibited visible SNR gains in the early stages, indicating the critical role of early denoising in restoring high-frequency details. ... Compared to the original SUPIR (blue), SUPIR+TDS (green) consistently achieves higher SNR at the same step count, even with fewer total steps, indicating its effectiveness in high-frequency signal restoration."

The TDS schedule (Eq. 3, Sec. 3.2) is deliberately constructed to concentrate samples exactly in the early (1000-700) and late (400-0) stages that Fig. 1 identifies as the phases where HF SNR improves. Fig. 7 then evaluates TDS using the same HF SNR metric: if you allocate iterations to the stages where a metric rises, that metric naturally rises. This is a self-consistency check, not an independent confirmation of the method's perceptual benefit. The no-reference metrics (NIQE, MUSIQ, QAlign) provide the real external validation, so the step is only partially circular.
fitted input called prediction [Sec. 4.1, A.2 (Table 5), A.7, A.9 (hyperparameter sensitivity)]
"Since each anchor method contains distinct backbones and base samplers, small differences in the values of the hyperparameters are taken to achieve the best performance. ... while the hyperparameters of the TSS strategy show robustness across various test datasets, they exhibit sensitivity to model architecture, as indicated in Tab. 10."

The paper's headline comparison (e.g., SUPIR_TSS with 7 steps outperforming ResShift-15 and SinSR-1 in Table 1) is obtained after per-model selection of the (a,n) hyperparameters on the same benchmarks reported in that table. The strategy is presented as a training-free method, but its quantitative output is the result of test-set tuning rather than a fixed, parameter-free rule. This is not a tautological reduction because the evaluation still uses external perceptual metrics and does not fit competitor outputs, but the magnitude of the reported improvement is partly a selection artifact.

full rationale

The paper's main derivation is an empirical observation about SUPIR's denoising HF SNR (Fig. 1), followed by a schedule (TDS, Eq. 3) that implements that observation and an optional spatial modulation (SDS, Eqs. 4-9). The schedule is not derived from a theorem; it is a heuristic justified by the same HF-SNR analysis that is later used to show benefit (Fig. 7), which is self-confirming rather than independent. The spatial component (SDS) changes the per-pixel time embedding of a pre-trained network, and A.9 concedes boundary artifacts from large neighboring timestep intervals; this is a correctness/mechanism concern, not a circularity by construction. No load-bearing self-citation or imported uniqueness theorem is present: related-work citations to the authors' prior papers (e.g., Qin et al. 2023, 2024) are background, not the basis of the claim. The final benchmark claims rest on external perceptual metrics (NIQE, CLIPIQA, MUSIQ, QAlign) and head-to-head comparisons against SinSR and ResShift, so the result is not a tautology. The main substantive circularity risk is that the per-model hyperparameters (Table 5) are tuned per dataset/model to maximize those same metrics, and the paper explicitly says small differences are taken to achieve the best performance. Thus the improvement reported in Table 1 is partly a fitted quantity, though competitor numbers are externally produced. Overall, the central claim has independent content and should not be rated highly circular; score 3 reflects the two self-referential elements (HF-SNR self-validation and benchmark-specific hyperparameter tuning) rather than a forced equivalence.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method is a schedule transformation and a set of inference-time modifications to a pretrained network. The main free parameters are the TDS shape parameters and the variance map computation, which are tuned per base model rather than derived from first principles.

free parameters (4)

TDS power factor n (and range nmin, nmax) = Per model: SUPIR n in [2.2,2.5], PASD n in [1,2], StableSR n in [1,1.2]
Controls the degree of non-uniformity of the time schedule. Tuned per anchor method in Tab. 5 to achieve best benchmark scores.
TDS transition point a (and range amin, amax) = Per model: SUPIR a in [0.58,0.63], PASD a in [0.4,0.6], StableSR a in [0.45,0.65]
Splits early and late stages. Also tuned per model in Tab. 5.
SDS linear mapping ranges (min/max for n and a over spatial positions) = Same as nmin, nmax, amin, amax values; applied via linear functions in Eq. 7
These ranges determine how strongly texture-rich regions are concentrated in early and late steps. They are not chosen independently from TDS ranges.
Variance computation field size (33x33) and Gaussian blur size = 33x33 field mentioned in Sec. 3.2, blur size not specified
Used to compute the smooth local variance map. The blur size is unspecified, making exact replication difficult.

assumptions (3)

domain assumption The final denoised result is a valid noise-free reference for SNR analysis of intermediate outputs.
Sec. 3.1 computes SNR of intermediate states relative to the final image. This assumes the final output is the ground truth for the denoising trajectory, which may overstate or understate recovery dynamics.
ad hoc to paper A pre-trained diffusion model can accept spatially varying per-pixel timestep embeddings without retraining and still produce coherent outputs.
This is required for SDS. The paper itself notes in A.9 that large neighboring timestep differences cause artifacts, so the assumption holds only approximately.
domain assumption Local variance of the low-resolution image is a good proxy for the amount of high-frequency texture content.
Used to assign different schedules per pixel in SDS. No evidence is provided beyond the qualitative observation in Fig. 1.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Accelerating Diffusion-based Super-Resolution with Dynamic Time-Spatial Sampling." pith.science (2026). https://pith.science/paper/PE636MZM

@misc{pith2026250512048,
  author       = {Pith},
  title        = {Pith review of: Accelerating Diffusion-based Super-Resolution with Dynamic Time-Spatial Sampling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PE636MZM}},
  note         = {Machine review of arXiv:2505.12048}
}

read the original abstract

Diffusion models have gained attention for their success in modeling complex distributions, achieving impressive perceptual quality in SR tasks. However, existing diffusion-based SR methods often suffer from high computational costs, requiring numerous iterative steps for training and inference. Existing acceleration techniques, such as distillation and solver optimization, are generally task-agnostic and do not fully leverage the specific characteristics of low-level tasks like super-resolution (SR). In this study, we analyze the frequency- and spatial-domain properties of diffusion-based SR methods, revealing key insights into the temporal and spatial dependencies of high-frequency signal recovery. Specifically, high-frequency details benefit from concentrated optimization during early and late diffusion iterations, while spatially textured regions demand adaptive denoising strategies. Building on these observations, we propose the Time-Spatial-aware Sampling strategy (TSS) for the acceleration of Diffusion SR without any extra training cost. TSS combines Time Dynamic Sampling (TDS), which allocates more iterations to refining textures, and Spatial Dynamic Sampling (SDS), which dynamically adjusts strategies based on image content. Extensive evaluations across multiple benchmarks demonstrate that TSS achieves state-of-the-art (SOTA) performance with significantly fewer iterations, improving MUSIQ scores by 0.2 - 3.0 and outperforming the current acceleration methods with only half the number of steps.

Figures

Figures reproduced from arXiv: 2505.12048 by the authors.

**Figure 1.** (a) SNR of different frequency components in SUPIR denoising, where high-frequency signals show a unique two-stage pattern. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 1.** a, SNR increases significantly in the late stages (400 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Overview of the proposed Time-Spatial-aware Sampling. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Figures from the paper (7 more)

**Figure 3.** Figure 3: Illustration of the Time Dynamic Sampling strategy. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]

**Figure 5.** Figure 5: Qualitative comparison with state-of-the-art methods. Top: real-world sample from RealPhoto60 datasets. Bottom: synthetic [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 7.** Figure 7: Comparison of SNR for high-frequency components at [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

**Figure 8.** Figure 8: Last 4 steps in PASD’s 7-step denoising process. ”+TDS” [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

**Figure 9.** Figure 9: More visual comparison with state-of-the-art Real-world SR methods. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

**Figure 10.** Figure 10: More visual comparison with state-of-the-art Real-world SR methods. [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

**Figure 11.** Figure 11: More visual comparison with state-of-the-art Real-world SR methods. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 31 canonical work pages

[1]

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

[Agustsson and Timofte, 2017] Eirikur Agustsson and Radu Timofte. Ntire 2017 challenge on single image super- resolution: Dataset and study. In The IEEE Conference on Computer Vision and Pattern Recognition Workshops ,

work page 2017
[7]

Denoising diffusion probabilistic models

[Ho et al., 2020] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, pages 6840–6851,

work page 2020
[10]

Elucidating the design space of diffusion-based generative models

[Karras et al., 2022] Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. Advances in neural in- formation processing systems, pages 26565–26577,

work page 2022
[12]

Musiq: Multi-scale im- age quality transformer

[Ke et al., 2021] Junjie Ke, Qifei Wang, Yilin Wang, Pey- man Milanfar, and Feng Yang. Musiq: Multi-scale im- age quality transformer. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5148– 5157,

work page 2021
[13]

Swinir: Image restoration using swin transformer

[Liang et al., 2021] Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. In Proceedings of the IEEE/CVF international conference on computer vi- sion, pages 1833–1844,

work page 2021
[14]

Reconstructed con- volution module based look-up tables for efficient image super-resolution

[Liu et al., 2023] Guandu Liu, Yukang Ding, Mading Li, Ming Sun, Xing Wen, and Bin Wang. Reconstructed con- volution module based look-up tables for efficient image super-resolution. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision , pages 12217– 12226,

work page 2023
[16]

completely blind

[Mittal et al., 2013] Anish Mittal, Rajiv Soundararajan, and Alan C. Bovik. Making a “completely blind” image qual- ity analyzer. IEEE Signal Processing Letters, pages 209– 212,

work page 2013
[18]

Blind image super-resolution with rich texture-aware codebook

[Qin et al., 2023] Rui Qin, Ming Sun, Fangyuan Zhang, Xing Wen, and Bin Wang. Blind image super-resolution with rich texture-aware codebook. In Proceedings of the 31st ACM International Conference on Multimedia , page 676–687,

work page 2023

Show all 42 references

[19]

A new dataset and framework for real-world blurred images super-resolution

[Qin et al., 2024] Rui Qin, Ming Sun, Chao Zhou, and Bin Wang. A new dataset and framework for real-world blurred images super-resolution. In European Conference on Computer Vision, pages 56–75,

2024
[20]

Xpsr: Cross- modal priors for diffusion-based image super-resolution

[Qu et al., 2025] Yunpeng Qu, Kun Yuan, Kai Zhao, Qizhi Xie, Jinhua Hao, Ming Sun, and Chao Zhou. Xpsr: Cross- modal priors for diffusion-based image super-resolution. In Computer Vision – ECCV 2024, pages 285–303,

2025
[21]

High-resolution image synthesis with latent diffusion models

[Rombach et al., 2022] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 10684–10695,

2022
[22]

Image super-resolution via iterative refinement

[Saharia et al., 2022] Chitwan Saharia, Jonathan Ho, William Chan, Tim Salimans, David J Fleet, and Mo- hammad Norouzi. Image super-resolution via iterative refinement. IEEE transactions on pattern analysis and machine intelligence, pages 4713–4726,

2022
[23]

Denoising diffusion implicit models

[Song et al., 2020] Jiaming Song, Chenlin Meng, and Ste- fano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502,

2020 arXiv
[24]

A performance evaluation of loss functions for deep face recognition, computer vision

[Srivastava et al., 2019] Y Srivastava, V Murali, and SR Dubey. A performance evaluation of loss functions for deep face recognition, computer vision. In Pattern Recognition, Image Processing, and Graphics: 7th National Conference, pages 22–24,

2019
[26]

Real-esrgan: Training real-world blind super-resolution with pure synthetic data

[Wang et al., 2021b] Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. In Pro- ceedings of the IEEE/CVF international conference on computer vision, pages 1905–1914,

1905
[27]

Exploring clip for assessing the look and feel of images

[Wang et al., 2023] Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Exploring clip for assessing the look and feel of images. In Proceedings of the AAAI Confer- ence on Artificial Intelligence, pages 2555–2563,

2023
[28]

Component divide-and-conquer for real-world image super-resolution

[Wei et al., 2020] Pengxu Wei, Ziwei Xie, Hannan Lu, Zongyuan Zhan, Qixiang Ye, Wangmeng Zuo, and Liang Lin. Component divide-and-conquer for real-world image super-resolution. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings...

2020
[29]

Q-align: Teaching lmms for visual scoring via discrete text-defined levels

[Wu et al., 2023] Haoning Wu, Zicheng Zhang, Weixia Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Yixuan Gao, Annan Wang, Erli Zhang, Wenxiu Sun, et al. Q-align: Teaching lmms for visual scoring via discrete text-defined levels. arXiv preprint arXiv:2312.17090,

2023 arXiv
[30]

One-step effective diffusion net- work for real-world image super-resolution.arXiv preprint arXiv:2406.08177,

[Wu et al., 2024a] Rongyuan Wu, Lingchen Sun, Zhiyuan Ma, and Lei Zhang. One-step effective diffusion net- work for real-world image super-resolution.arXiv preprint arXiv:2406.08177,

arXiv
[31]

Star: Spatial-temporal augmentation with text-to-video mod- els for real-world video super-resolution

[Xie et al., 2025] Rui Xie, Yinhong Liu, Penghao Zhou, Chen Zhao, Jun Zhou, Kai Zhang, Zhenyu Zhang, Jian Yang, Zhenheng Yang, and Ying Tai. Star: Spatial-temporal augmentation with text-to-video mod- els for real-world video super-resolution. arXiv preprint arXiv:2501.02976,

2025 arXiv
[32]

Accelerating diffusion sampling with optimized time steps

[Xue et al., 2024] Shuchen Xue, Zhaoqiang Liu, Fei Chen, Shifeng Zhang, Tianyang Hu, Enze Xie, and Zhenguo Li. Accelerating diffusion sampling with optimized time steps. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8292– 8301,

2024
[33]

Pixel-aware stable diffu- sion for realistic image super-resolution and personalized stylization

[Yang et al., 2025] Tao Yang, Rongyuan Wu, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffu- sion for realistic image super-resolution and personalized stylization. In European Conference on Computer Vision, pages 74–91,

2025
[34]

One-step diffusion with distribution matching distillation

[Yin et al., 2024] Tianwei Yin, Micha ¨el Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, ...

2024
[35]

Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild

[Yu et al., 2024] Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, and Chao Dong. Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild. In Proceedings of the IEEE/CVF Conference on Co...

2024
[36]

Resshift: Efficient diffusion model for image super-resolution by residual shifting

[Yue et al., 2024] Zongsheng Yue, Jianyi Wang, and Chen Change Loy. Resshift: Efficient diffusion model for image super-resolution by residual shifting. Advances in Neural Information Processing Systems,

2024
[37]

The unreasonable effectiveness of deep features as a perceptual metric

[Zhang et al., 2018] Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595,

2018
[39]

Adding conditional control to text-to-image dif- fusion models

[Zhang et al., 2023] Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image dif- fusion models. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pages 3836–3847,

2023
[40]

Quality-aware pre-trained mod- els for blind image quality assessment

[Zhao et al., 2023] Kai Zhao, Kun Yuan, Ming Sun, Mad- ing Li, and Xing Wen. Quality-aware pre-trained mod- els for blind image quality assessment. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 22302–22313,

2023
[41]

Unipc: A unified predictor- corrector framework for fast sampling of diffusion mod- els

[Zhao et al., 2024] Wenliang Zhao, Lujia Bai, Yongming Rao, Jie Zhou, and Jiwen Lu. Unipc: A unified predictor- corrector framework for fast sampling of diffusion mod- els. Advances in Neural Information Processing Systems,

2024
[42]

Dip-go: A diffusion pruner via few-step gradient optimization

[Zhu et al., 2024] Haowei Zhu, Dehua Tang, Ji Liu, Mingjie Lu, Jintu Zheng, Jinzhang Peng, Dong Li, Yu Wang, Fan Jiang, Lu Tian, et al. Dip-go: A diffusion pruner via few-step gradient optimization. arXiv preprint arXiv:2410.16942, 2024

2024 arXiv
[1967]

One step diffusion- based super-resolution with time-aware distillation

[He et al., 2024] Xiao He, Huaao Tang, Zhijun Tu, Junchao Zhang, Kun Cheng, Hanting Chen, Yong Guo, Mingrui Zhu, Nannan Wang, Xinbo Gao, et al. One step diffusion- based super-resolution with time-aware distillation. arXiv preprint arXiv:2408.07476,

2024 arXiv
[2010]

Real-world super- resolution via kernel estimation and noise injection

[Ji et al., 2020] Xiaozhong Ji, Yun Cao, Ying Tai, Chengjie Wang, Jilin Li, and Feiyue Huang. Real-world super- resolution via kernel estimation and noise injection. In proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition workshops , pages 466–467,

2020
[2013]

Sdxl: Improving la- tent diffusion models for high-resolution image synthesis

[Podell et al., 2023] Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving la- tent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952,

2023 arXiv
[2017]

Plug-and-play tri-branch invertible block for image rescaling

[Bao et al., 2025] Jingwei Bao, Jinhua Hao, Pengcheng Xu, Ming Sun, Chao Zhou, and Shuyuan Zhu. Plug-and-play tri-branch invertible block for image rescaling. In Pro- ceedings of the AAAI Conference on Artificial Intelligence, pages 1826–1834,

2025
[2018]

Designing a practical degradation model for deep blind image super-resolution

[Zhang et al., 2021] Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timofte. Designing a practical degradation model for deep blind image super-resolution. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4791–4800,

2021
[2019]

Bovik, H.R

[Wang et al., 2004] Zhou 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, pages 600–612,

2004
[2020]

Im- age quality metrics: Psnr vs

[Hor´e and Ziou, 2010] Alain Hor ´e and Djemel Ziou. Im- age quality metrics: Psnr vs. ssim. In 2010 20th Inter- national Conference on Pattern Recognition, pages 2366– 2369,

2010
[2021]

Cochran, J.W

[Cochran et al., 1967] W.T. Cochran, J.W. Cooley, D.L. Favin, H.D. Helms, R.A. Kaenel, W.W. Lang, G.C. Ma- ling, D.E. Nelson, C.M. Rader, and P.D. Welch. What is the fast fourier transform? Proceedings of the IEEE, pages 1664–1674,

1967
[2022]

Denoising diffusion restora- tion models

[Kawar et al., 2022] Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restora- tion models. Advances in Neural Information Processing Systems, pages 23593–23606,

2022
[2023]

Deepcache: Accelerating diffusion models for free

[Ma et al., 2024] Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 15762– 15772,

2024
[2024]

Ilvr: Conditioning method for denoising diffusion probabilistic models

[Choi et al., 2021] Jooyoung Choi, Sungwon Kim, Yonghyun Jeong, Youngjune Gwon, and Sungroh Yoon. Ilvr: Conditioning method for denoising diffusion probabilistic models. arXiv preprint arXiv:2108.02938 ,

2021 arXiv
[2025]

Cassr: Activating image power for real-world image super- resolution

[Chen et al., 2024] Haolan Chen, Jinhua Hao, Kai Zhao, Kun Yuan, Ming Sun, Chao Zhou, and Wei Hu. Cassr: Activating image power for real-world image super- resolution. arXiv preprint arXiv:2403.11451,

2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

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

[Agustsson and Timofte, 2017] Eirikur Agustsson and Radu Timofte. Ntire 2017 challenge on single image super- resolution: Dataset and study. In The IEEE Conference on Computer Vision and Pattern Recognition Workshops ,

work page 2017

[2] [7]

Denoising diffusion probabilistic models

[Ho et al., 2020] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, pages 6840–6851,

work page 2020

[3] [10]

Elucidating the design space of diffusion-based generative models

[Karras et al., 2022] Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. Advances in neural in- formation processing systems, pages 26565–26577,

work page 2022

[4] [12]

Musiq: Multi-scale im- age quality transformer

[Ke et al., 2021] Junjie Ke, Qifei Wang, Yilin Wang, Pey- man Milanfar, and Feng Yang. Musiq: Multi-scale im- age quality transformer. In Proceedings of the IEEE/CVF international conference on computer vision, pages 5148– 5157,

work page 2021

[5] [13]

Swinir: Image restoration using swin transformer

[Liang et al., 2021] Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration using swin transformer. In Proceedings of the IEEE/CVF international conference on computer vi- sion, pages 1833–1844,

work page 2021

[6] [14]

Reconstructed con- volution module based look-up tables for efficient image super-resolution

[Liu et al., 2023] Guandu Liu, Yukang Ding, Mading Li, Ming Sun, Xing Wen, and Bin Wang. Reconstructed con- volution module based look-up tables for efficient image super-resolution. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision , pages 12217– 12226,

work page 2023

[7] [16]

completely blind

[Mittal et al., 2013] Anish Mittal, Rajiv Soundararajan, and Alan C. Bovik. Making a “completely blind” image qual- ity analyzer. IEEE Signal Processing Letters, pages 209– 212,

work page 2013

[8] [18]

Blind image super-resolution with rich texture-aware codebook

[Qin et al., 2023] Rui Qin, Ming Sun, Fangyuan Zhang, Xing Wen, and Bin Wang. Blind image super-resolution with rich texture-aware codebook. In Proceedings of the 31st ACM International Conference on Multimedia , page 676–687,

work page 2023