Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Fine-structure Preserved Real-world Image Super-resolution via Transfer VAE Training

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A transfer-trained 4x VAE keeps Stable Diffusion's latent space while preserving fine text and textures in one-step real-world super-resolution.

desk verdict A practical two-stage VAE-transfer recipe for SD-based super-resolution, with honest ablations and one unmeasured claim (latent alignment) that needs a direct test. read the letter →

arxiv 2507.20291 v1 pith:FQOTCVTD submitted 2025-07-27 cs.CV

Qiaosi Yi , Shuai Li , Rongyuan Wu , Lingchen Sun , Yuhui Wu , Lei Zhang This is my paper
classification cs.CV
keywords TransferVAETrainingreal-worldimagesuper-resolutionStableDiffusionlatentspacealignmentfine-structurepreservationone-stepscenetextcompute-efficientUNet
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 argues that the main reason diffusion-based real-world super-resolution methods blur small text and fine textures is the 8x spatial compression of the Stable Diffusion VAE, and proposes a way to replace it with a 4x VAE without retraining the pretrained denoising UNet. The proposed Transfer VAE Training (TVT) first trains a compact 4x decoder on latents produced by the original 8x encoder, then trains a compact 4x encoder against that fixed decoder. The result is an adapted VAE-D4 whose latent space stays compatible with the pretrained UNet while reconstructing much sharper details: on Urban100 its reconstruction PSNR is 32.35 dB versus 25.29 dB for VAE-D8. Combined with a compute-efficient UNet that processes 128x128 features only in replicated first and last layers, the method restores fine structures in one step with 1.97 T FLOPs, fewer than the compared one-step diffusion models.

What carries the argument

The central object is the transfer-trained VAE pair (E4, D4), created by fixing the old 8x encoder E8 and training D4 to invert its latents, with the input first upsampled 2x so the resolutions match, and then fixing D4 and training E4 with reconstruction losses only, L1 and LPIPS. That two-stage coupling is what transfers latent-space alignment, letting the pretrained UNet be reused without retraining from scratch. The supporting mechanism is a compute-efficient UNet in which the first and last layers are replicated, initialized from the pretrained weights, and fully fine-tuned to handle 128x128 features while the rest of the UNet is adapted with LoRA on 64x64 features.

What would settle it

Compute a held-out set of images, encode each with E8 after 2x upsampling and with E4, and measure a distributional divergence between the two latent sets, such as per-channel means and variances, sliced Wasserstein distance, or a classifier trained to tell them apart. If the divergence is large while downstream gains persist, the latent-alignment story is wrong and the gains instead come from LoRA or UNet adaptation compensating for a mismatched latent space. Conversely, the claim weakens if, even with matched latents, fine-structure gains vanish when the UNet is frozen and only the VAE components are swapped.

Watch

Extended reading notes

Core claim

The paper's central claim is that a two-stage transfer training schedule can convert Stable Diffusion's 8x-downsampled VAE into a 4x-downsampled VAE that plugs into the pretrained UNet: stage one trains the new decoder to invert the old encoder's latents, stage two trains the new encoder to feed that fixed decoder. This keeps the modified latent space close enough to the original that the pretrained diffusion prior keeps working, while the gentler 4x compression retains high-frequency structure that the 8x VAE loses. The paper reports consistent gains over multi-step and one-step SD-based super-resolution methods on RealSR, DRealSR, and DIV2K-val, and on the scene-text benchmark RealCE shows higher character recognition accuracy and normalized edit distance. It also demonstrates the efficiency story: a compact three-stage VAE-D4 and a UNet that only fine-tunes replicated first and last layers at 128x128 resolution yield 1.97 T FLOPs, below the one-step baselines.

Load-bearing premise

The load-bearing assumption is that training the new 4x encoder with only reconstruction losses through the fixed new decoder produces latents whose distribution is close enough to the original 8x VAE's latents for the pretrained UNet to work; the paper demonstrates this through downstream super-resolution quality rather than by measuring the distance between the two latent distributions.

Editorial extensions

If this is right

  • SD-based Real-ISR pipelines can adopt lower-compression VAEs without retraining the UNet from scratch, raising fidelity metrics while keeping the diffusion prior.
  • One-step real-world super-resolution can preserve legible small text, as reflected in higher recognition accuracy and normalized edit distance on the scene-text benchmark.
  • Because only the replicated UNet layers are fully trained and the rest is LoRA-adapted, the method inherits the pretrained diffusion prior, which is useful in data-limited restoration settings.
  • The compact VAE-D4 reduces parameters by 81.89% and FLOPs by 38.89% relative to VAE-D8, pointing toward lower-cost deployment of fine-structure-preserving super-resolution.

Reading between the lines

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

  • The latent-alignment claim rests on behavioral evidence; an explicit distributional distance between E4 latents and E8 latents, such as sliced Wasserstein distance or per-channel statistics on a held-out set, would test whether the alignment is real or whether LoRA adaptation compensates for a shifted latent space.
  • A testable extension is to apply the same two-stage decoder-first, encoder-second transfer to other downsampling factors or other base diffusion VAEs, to see whether TVT is a generic recipe for swapping compression ratios.
  • The fine-structure gains could be partly attributable to the higher reconstruction quality of VAE-D4 rather than to latent alignment; an ablation that freezes the UNet and swaps only the VAE components would separate these contributions.
  • Since reconstruction metrics are reported only on Urban100, broader evaluation on text-heavy and texture-heavy datasets would clarify whether the fine-structure advantage persists outside the tested benchmarks.
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 / 5 minor

Summary. The paper proposes Transfer VAE Training (TVT), a two-stage procedure that adapts the 8x-downsampling VAE of Stable Diffusion 2.1 into a compact 4x-downsampling VAE while attempting to keep the latent space compatible with the pretrained UNet. Stage 1 trains a compact 4x decoder on latents produced by the original 8x encoder; Stage 2 trains a 4x encoder against the fixed decoder using only reconstruction losses. The authors also introduce a compute-efficient UNet that replicates the first and last pretrained UNet layers to process the higher-resolution latents. The resulting one-step real-world super-resolution system is evaluated on RealSR, DRealSR, DIV2K-val, and the RealCE scene-text benchmark, reporting better fine-structure preservation and lower FLOPs than several one-step and multi-step SD-based baselines. The paper includes ablations for the VAE architecture, the TVT strategy, and the compute-efficient UNet, and it provides a public code link.

Significance. If the central claim holds, TVT is a practical contribution: it replaces the 8x SD VAE with a 4x VAE while reusing the pretrained UNet, improving text and detail restoration in one-step real-world super-resolution at reduced cost. The paper has concrete strengths: it ships code, reports FLOPs and parameter counts for all compared methods, evaluates on a dedicated text super-resolution benchmark with recognition metrics, and includes ablations that isolate the contributions of the compact VAE, the TVT strategy, and the compute-efficient UNet. The main risk is that the load-bearing premise of latent-space alignment is asserted rather than directly measured, and the reported quantitative gains are not accompanied by error bars or significance tests.

major comments (3)
  1. [Sec. 3.2, Eq. (4), Table 6] The paper's central claim is that TVT aligns the new VAE-D4 latent space with the original VAE-D8 latent space so that the pretrained UNet can be reused. However, the Stage-2 objective in Eq. (4) only minimizes L1 and LPIPS reconstruction losses through the fixed decoder D4; it contains no term that penalizes divergence between E4(I) and E8(I↑2), and no distance between these latent distributions is measured anywhere in the paper. Since D4 is many-to-one, D4(E4(I)) ≈ I does not imply E4(I) lies in the E8-latent distribution. The ablation T2 in Table 6, which explicitly aligns latents with L1 loss and performs worse than TVT, is consistent with a latent-space mismatch that is later compensated by LoRA fine-tuning rather than by true alignment. To support the transfer claim, the authors should report a direct latent-space distance (e.g., FID/MMD between E4(I) and E8(I↑2) latent sets) and include a no-LoRA or LoRA-controlled experiment showing that downstream gains do not come solely from the adaptation layers.
  2. [Sec. 4.1, Testing Datasets] The paper states that the DIV2K-val dataset consists of 3,000 images of 512x512 resolution. The public DIV2K validation set contains 100 images, and 3,000 is not a standard count for that benchmark. If the authors used a different test set, such as a degraded version with 3,000 patches or a custom crop, they must specify the exact generation protocol and source; otherwise the experimental results on this dataset are not reproducible and the comparison in Table 2 cannot be verified by other researchers.
  3. [Sec. 4.2, Table 2] The quantitative comparisons are reported as single numbers without error bars, confidence intervals, or significance tests. Some of the headline differences, such as SSIM 0.7596 vs 0.7341 on RealSR and LPIPS 0.2587 vs 0.2921, may be meaningful, but the paper uses the phrase 'significantly higher' without statistical support. Given that many compared methods are also stochastic or have multiple checkpoints, the authors should either provide variance across multiple runs or temper the significance language.
minor comments (5)
  1. [Sec. 4.2, Qualitative Comparison] The phrase 'The visaul comparisons are shown in Fig. 4' contains a typo: 'visaul' should be 'visual'.
  2. [Sec. 3.3, Eq. (5)] The notation in Eq. (5) is unclear: z is introduced but zLR is used in the preceding sentence, and the relationship between z, t, and the one-step sampling formula should be stated explicitly so the reader can reproduce the inference procedure.
  3. [Sec. 4.3, Ablation Study] In the paragraph on the TVT strategy, the sentence 'This highlights the effectiveness of the TVT strategy in aligning the the latent spaces' contains a duplicated 'the'.
  4. [Sec. 3.2, Compact VAE-D4 Design] The architecture description says the compact VAE-D4 uses channel sizes 128/256/256, but Table 5 describes V1 and V2 as using 128/256/512 channels and TVT as the final configuration. The relationship between these channel counts should be clarified, and the exact stage configuration of the final VAE-D4 should be given in a table or figure.
  5. [Table 4] The FLOPs comparison does not state the input resolution and whether FLOPs are measured for the entire SR pipeline or only the diffusion UNet. This should be specified so that the 1.97T figure can be reproduced.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TVT is an empirical system evaluated on external benchmarks; the latent-alignment claim is an unsupported assumption, not a circular derivation.

full rationale

The derivation chain is not circular. Stage 1 trains the VAE-D4 decoder on VAE-D8 encoder latents with reconstruction and GAN losses (Eqs. 1-2), and stage 2 trains the VAE-D4 encoder through the fixed decoder with L1+LPIPS losses (Eqs. 3-4). The paper's central claim that this aligns the new encoder-decoder with the original latent space is not proven by those equations—no latent-distance measurement is reported, and a many-to-one decoder does not force distributional alignment—but this is a missing-support or correctness concern, not a case of a prediction reducing by construction to its inputs. No fitted constant is renamed as a prediction; the Real-ISR results are evaluated on external benchmarks (RealSR, DRealSR, DIV2K-val, RealCE), and the ablation T2, which directly enforces L1 latent alignment and performs worse, is evidence against the stated mechanism rather than evidence of circularity. Self-citations to OSEDiff and SeeSR are baseline methods and the LoRA/VSD training recipe is adopted as prior work, not as proof of TVT's own contribution. The paper is therefore self-contained against external validation, and no step reduces to its own definition.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

The method rests on several domain assumptions about how latent diffusion priors transfer. The most fragile are the two ad hoc assumptions that a 4x autoencoder can approximate the 8x latent distribution without direct distributional supervision, and that reconstruction-only training of the encoder is sufficient to achieve latent-space compatibility.

free parameters (5)
  • LPIPS loss weight lambda1 = 2
    Hand-chosen in Eq. 7; no sensitivity analysis is reported. It directly affects the fidelity-perceptual balance of the SR result.
  • VSD loss weight lambda2 = 1
    Hand-chosen in Eq. 7; no sensitivity analysis is reported. It controls the influence of variational score distillation on one-step training.
  • LoRA rank = 4
    Hand-chosen in Sec. 4.1; sets the adaptation capacity of the UNet during fine-tuning.
  • Compact VAE-D4 stage/channel configuration = 3 stages, channels 128/256/256, no middle stage
    Hand-pruned from VAE-D8 in Sec. 3.2; ablation V1/V2 shows similar downstream performance, but the exact configuration is an ad hoc design choice.
  • Adaptive GAN loss weight lambda_D = gradient-norm ratio
    Defined in Eq. 2 as a gradient-based adaptive weighting; a heuristic design choice for decoder training.
assumptions (6)
  • domain assumption The pre-trained SD 2.1 UNet retains a usable generative prior for real-world SR when the VAE latent space is approximately preserved.
    Invoked throughout Sec. 3.2 and 3.3; the entire TVT design exists to avoid retraining the UNet.
  • ad hoc to paper A 4x VAE with 4 latent channels can approximate the 8x VAE's latent distribution closely enough for the pretrained UNet.
    Central premise of TVT; no direct latent-distribution distance is measured, only downstream SR metrics (Sec. 3.2, Eq. 4).
  • ad hoc to paper Training D4 on E8 outputs of 2x-upsampled images transfers to E4-produced latents.
    Stage-1 decoder sees only E8 latents; stage-2 encoder is trained by reconstruction through D4, not by matching E8 latents (Sec. 3.2).
  • domain assumption Real-ESRGAN degradation pipeline adequately simulates real-world degradations for training.
    Used to generate LR-HR pairs for Real-ISR training (Sec. 4.1).
  • domain assumption Selected reference and no-reference metrics reflect fine-structure preservation.
    The fine-structure claim is operationalized through PSNR, SSIM, LPIPS, CLIPIQA, MUSIQ, and text accuracy metrics (Sec. 4.1).
  • domain assumption LoRA tuning of the original UNet plus full tuning of replicated first/last layers preserves pretrained diffusion priors while adapting to 128x128 latents.
    Used in Sec. 3.3 and supported only by ablations S4/S5/TVT, not by direct prior-integrity measurements.
invented entities (2)
  • VAE-D4 (4x downsampled VAE variant) independent evidence
    purpose: Higher-fidelity latent encoding and decoding that preserves fine structure at 4x compression while remaining compatible with the SD UNet.
    Supported by Table 1 reconstruction metrics and ablation S4 vs S1; an official code repo is announced.
  • CE-UNet (compute-efficient UNet with replicated first/last layers) independent evidence
    purpose: Processes 128x128 latent features from VAE-D4 with fewer FLOPs than a full 128x128 UNet, reusing the pretrained 64x64 UNet.
    Supported by Table 4 FLOPs and ablation TVT vs S4; the architecture is described with enough detail to implement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fine-structure Preserved Real-world Image Super-resolution via Transfer VAE Training." pith.science (2026). https://pith.science/paper/FQOTCVTD

@misc{pith2026250720291,
  author       = {Pith},
  title        = {Pith review of: Fine-structure Preserved Real-world Image Super-resolution via Transfer VAE Training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FQOTCVTD}},
  note         = {Machine review of arXiv:2507.20291}
}
abstract

Impressive results on real-world image super-resolution (Real-ISR) have been achieved by employing pre-trained stable diffusion (SD) models. However, one critical issue of such methods lies in their poor reconstruction of image fine structures, such as small characters and textures, due to the aggressive resolution reduction of the VAE (eg., 8$\times$ downsampling) in the SD model. One solution is to employ a VAE with a lower downsampling rate for diffusion; however, adapting its latent features with the pre-trained UNet while mitigating the increased computational cost poses new challenges. To address these issues, we propose a Transfer VAE Training (TVT) strategy to transfer the 8$\times$ downsampled VAE into a 4$\times$ one while adapting to the pre-trained UNet. Specifically, we first train a 4$\times$ decoder based on the output features of the original VAE encoder, then train a 4$\times$ encoder while keeping the newly trained decoder fixed. Such a TVT strategy aligns the new encoder-decoder pair with the original VAE latent space while enhancing image fine details. Additionally, we introduce a compact VAE and compute-efficient UNet by optimizing their network architectures, reducing the computational cost while capturing high-resolution fine-scale features. Experimental results demonstrate that our TVT method significantly improves fine-structure preservation, which is often compromised by other SD-based methods, while requiring fewer FLOPs than state-of-the-art one-step diffusion models. The official code can be found at https://github.com/Joyies/TVT.

Figures

Figures reproduced from arXiv: 2507.20291 by the authors.

Figure 1
Figure 1. (a) Comparison of the reconstruction results of orig [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Illustration of our transfer VAE training strategy. We first train a VAE-D4 decoder conditioned on the original VAE-D8’s [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The structure of (a) VAE-D8 used in SD 2.1-base and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visual comparison with SD-based Real-ISR methods. Please zoom in for a better view. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visual comparison with STSR methods. than OSEDiff. Overall, our TVT strategy, together with ef￾ficient design of VAE-D4 and denoising UNet, makes our Real-ISR model effective and efficient. 4.3. Ablation Study Since it is time-consuming to train a VAE, in the ablation …
Figure 6
Figure 6. Figure 6: Our method faces challenges in recovering details when [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. VOSR: A Vision-Only Generative Model for Image Super-Resolution

    cs.CV 2026-04 conditional novelty 7.0 of 10

    VOSR shows that competitive generative image super-resolution with faithful structures can be achieved by training a diffusion-style model from scratch on visual data alone, using a vision encoder for guidance and a r...

Reference graph

Works this paper leans on

91 extracted references · 42 canonical work pages · cited by 1 Pith paper

  1. [1]

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

    Eirikur Agustsson and Radu Timofte. Ntire 2017 challenge on single image super-resolution: Dataset and study. In Pro- ceedings of the IEEE conference on computer vision and pat- tern recognition workshops, pages 126–135, 2017. 5

    2017

  2. [2]

    DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation

    Yuang Ai, Xiaoqiang Zhou, Huaibo Huang, Xiaotian Han, Zhengyu Chen, Quanzeng You, and Hongxia Yang. Dreamclear: High-capacity real-world image restora- tion with privacy-safe dataset curation. arXiv preprint arXiv:2410.18666, 2024. 2

    work page Pith review arXiv

  3. [3]

    Toward real-world single image super-resolution: A new benchmark and a new model

    Jianrui Cai, Hui Zeng, Hongwei Yong, Zisheng Cao, and Lei Zhang. Toward real-world single image super-resolution: A new benchmark and a new model. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 3086–3095, 2019. 2, 5

    2019

  4. [4]

    Adversarial diffusion compression for real-world image super-resolution

    Bin Chen, Gehui Li, Rongyuan Wu, Xindong Zhang, Jie Chen, Jian Zhang, and Lei Zhang. Adversarial diffusion compression for real-world image super-resolution. arXiv preprint arXiv:2411.13383, 2024. 2, 5

    arXiv 2024

  5. [5]

    Real-world blind super-resolution via feature matching with implicit high- resolution priors

    Chaofeng Chen, Xinyu Shi, Yipeng Qin, Xiaoming Li, Xi- aoguang Han, Tao Yang, and Shihui Guo. Real-world blind super-resolution via feature matching with implicit high- resolution priors. In Proceedings of the 30th ACM Interna- tional Conference on Multimedia , pages 1329–1338, 2022. 2

    2022

  6. [6]

    Topiq: A top-down approach from semantics to distortions for image quality assessment

    Chaofeng Chen, Jiadi Mo, Jingwen Hou, Haoning Wu, Liang Liao, Wenxiu Sun, Qiong Yan, and Weisi Lin. Topiq: A top-down approach from semantics to distortions for image quality assessment. IEEE Transactions on Image Processing,

  7. [7]

    Generalized and efficient 2d gaussian splatting for arbitrary- scale super-resolution

    Du Chen, Liyi Chen, Zhengqiang Zhang, and Lei Zhang. Generalized and efficient 2d gaussian splatting for arbitrary- scale super-resolution. arXiv preprint arXiv:2501.06838 ,

    arXiv

  8. [8]

    Toward gen- eralized image quality assessment: Relaxing the perfect ref- erence quality assumption

    Du Chen, Tianhe Wu, Kede Ma, and Lei Zhang. Toward gen- eralized image quality assessment: Relaxing the perfect ref- erence quality assumption. In Proceedings of the Computer Vision and Pattern Recognition Conference , pages 12742– 12752, 2025. 5

    2025

Show all 91 references
  1. [9]

    Pre-trained image processing transformer

    Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12299–12310, 2021. 2

    2021

  2. [10]

    Scene text tele- scope: Text-focused scene image super-resolution

    Jingye Chen, Bin Li, and Xiangyang Xue. Scene text tele- scope: Text-focused scene image super-resolution. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12026–12035, 2021. 5, 6

    2021

  3. [11]

    Activating more pixels in image super- resolution transformer

    Xiangyu Chen, Xintao Wang, Jiantao Zhou, Yu Qiao, and Chao Dong. Activating more pixels in image super- resolution transformer. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 22367–22377, 2023. 2

    2023

  4. [12]

    Dual aggregation transformer for image super-resolution

    Zheng Chen, Yulun Zhang, Jinjin Gu, Linghe Kong, Xi- aokang Yang, and Fisher Yu. Dual aggregation transformer for image super-resolution. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12312– 12321, 2023

    2023

  5. [13]

    Feedback network for mutually boosted stereo image super-resolution and disparity estimation

    Qinyan Dai, Juncheng Li, Qiaosi Yi, Faming Fang, and Guixu Zhang. Feedback network for mutually boosted stereo image super-resolution and disparity estimation. InProceed- ings of the 29th ACM international conference on multime- dia, pages 1985–1993, 2021

    1985

  6. [14]

    Second-order attention network for single im- age super-resolution

    Tao Dai, Jianrui Cai, Yongbing Zhang, Shu-Tao Xia, and Lei Zhang. Second-order attention network for single im- age super-resolution. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 11065–11074, 2019. 1, 2

    2019

  7. [15]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 2

    2021

  8. [16]

    Image quality assessment: Unifying structure and texture similarity

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P Simoncelli. Image quality assessment: Unifying structure and texture similarity. IEEE transactions on pattern analysis and ma- chine intelligence, 44(5):2567–2581, 2020. 5

    2020

  9. [17]

    Learning a deep convolutional network for image super-resolution

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

    2014

  10. [18]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. In Forty-first international conference on mach...

  11. [19]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014. 1

    2014

  12. [20]

    Blind super-resolution with iterative kernel correction

    Jinjin Gu, Hannan Lu, Wangmeng Zuo, and Chao Dong. Blind super-resolution with iterative kernel correction. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 1604–1613, 2019. 2

    2019

  13. [21]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 4

    2016

  14. [22]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 2, 5

    2020

  15. [23]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 3, 5

    2021 arXiv

  16. [24]

    Single image super-resolution from transformed self-exemplars

    Jia-Bin Huang, Abhishek Singh, and Narendra Ahuja. Single image super-resolution from transformed self-exemplars. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5197–5206, 2015. 4

    2015

  17. [25]

    Percep- tual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Percep- tual losses for real-time style transfer and super-resolution. In Computer Vision–ECCV 2016: 14th European Confer- ence, Amsterdam, The Netherlands, October 11-14, 2016, 9 Proceedings, Part II 14, pages 694–711. Springer...

    2016

  18. [26]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 4401–4410, 2019. 5

    2019

  19. [27]

    Snips: Solving noisy inverse problems stochastically

    Bahjat Kawar, Gregory Vaksman, and Michael Elad. Snips: Solving noisy inverse problems stochastically. Advances in Neural Information Processing Systems , 34:21757–21769,

  20. [28]

    Denoising diffusion restoration models

    Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restoration models. Advances in Neural Information Processing Systems, 35:23593–23606,

  21. [29]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5148–5157, 2021. 5

    2021

  22. [30]

    Classsr: A general framework to accelerate super- resolution networks by data characteristic

    Xiangtao Kong, Hengyuan Zhao, Yu Qiao, and Chao Dong. Classsr: A general framework to accelerate super- resolution networks by data characteristic. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12016–12025, 2021. 1

    2021

  23. [31]

    Reflash dropout in image super-resolution

    Xiangtao Kong, Xina Liu, Jinjin Gu, Yu Qiao, and Chao Dong. Reflash dropout in image super-resolution. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6002–6012, 2022. 1

    2022

  24. [32]

    The open images dataset v4: Unified image classification, object detection, and visual relationship detection at scale

    Alina Kuznetsova, Hassan Rom, Neil Alldrin, Jasper Ui- jlings, Ivan Krasin, Jordi Pont-Tuset, Shahab Kamali, Stefan Popov, Matteo Malloci, Alexander Kolesnikov, et al. The open images dataset v4: Unified image classification, object detection, and visual relationship detection...

    1956

  25. [33]

    Black Forest Labs. Flux. https://github.com/ black-forest-labs/flux, 2024. 3

    2024

  26. [34]

    Photo- realistic single image super-resolution using a generative ad- versarial network

    Christian Ledig, Lucas Theis, Ferenc Husz´ar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo- realistic single image super-resolution using a generative ad- versarial network. In Proceedings of the IE...

  27. [35]

    Multi-scale residual network for image super-resolution

    Juncheng Li, Faming Fang, Kangfu Mei, and Guixu Zhang. Multi-scale residual network for image super-resolution. In Proceedings of the European conference on computer vision (ECCV), pages 517–532, 2018. 2

    2018

  28. [36]

    Lsdir: A large scale dataset for image restoration

    Yawei Li, Kai Zhang, Jingyun Liang, Jiezhang Cao, Ce Liu, Rui Gong, Yulun Zhang, Hao Tang, Yun Liu, Denis Deman- dolx, et al. Lsdir: A large scale dataset for image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1775–17...

    2023

  29. [37]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. InProceedings of the IEEE/CVF inter- national conference on computer vision , pages 1833–1844,

  30. [38]

    Details or artifacts: A locally discriminative learning approach to realistic im- age super-resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Details or artifacts: A locally discriminative learning approach to realistic im- age super-resolution. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 5657–5666, 2022. 1, 2

    2022

  31. [39]

    Efficient and degradation-adaptive network for real-world image super- resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Efficient and degradation-adaptive network for real-world image super- resolution. In European Conference on Computer Vision , pages 574–591. Springer, 2022. 2

    2022

  32. [40]

    Ntire 2024 restore any im- age model (raim) in the wild challenge

    Jie Liang, Radu Timofte, Qiaosi Yi, Shuaizheng Liu, Lingchen Sun, Rongyuan Wu, Xindong Zhang, Hui Zeng, Lei Zhang, Yibin Huang, et al. Ntire 2024 restore any im- age model (raim) in the wild challenge. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2024

  33. [41]

    Enhanced deep residual networks for single image super-resolution

    Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. Enhanced deep residual networks for single image super-resolution. In Proceedings of the IEEE confer- ence on computer vision and pattern recognition workshops, pages 136–144, 2017. 1, 2

    2017

  34. [42]

    Diffbir: Towards blind image restoration with generative diffusion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Ben Fei, Bo Dai, Wanli Ouyang, Yu Qiao, and Chao Dong. Diffbir: Towards blind image restoration with generative diffusion prior. arXiv preprint arXiv:2308.15070, 2023. 1, 2, 5

    2023 arXiv

  35. [43]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6

    2017 arXiv

  36. [44]

    A text atten- tion network for spatial deformation robust scene text im- age super-resolution

    Jianqi Ma, Zhetong Liang, and Lei Zhang. A text atten- tion network for spatial deformation robust scene text im- age super-resolution. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 5911–5920, 2022. 5, 6

    2022

  37. [45]

    Text prior guided scene text image super-resolution

    Jianqi Ma, Shi Guo, and Lei Zhang. Text prior guided scene text image super-resolution. IEEE Transactions on Image Processing, 32:1341–1353, 2023. 5, 6

    2023

  38. [46]

    A benchmark for chinese-english scene text im- age super-resolution

    Jianqi Ma, Zhetong Liang, Wangmeng Xiang, Xi Yang, and Lei Zhang. A benchmark for chinese-english scene text im- age super-resolution. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 19452– 19461, 2023. 5, 6

    2023

  39. [47]

    Codi: Conditional diffusion distillation for higher-fidelity and faster image generation

    Kangfu Mei, Mauricio Delbracio, Hossein Talebi, Zhengzhong Tu, Vishal M Patel, and Peyman Milanfar. Codi: Conditional diffusion distillation for higher-fidelity and faster image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p...

    2024

  40. [48]

    The power of context: How multimodality improves image super- resolution

    Kangfu Mei, Hossein Talebi, Mojtaba Ardakani, Vishal M Patel, Peyman Milanfar, and Mauricio Delbracio. The power of context: How multimodality improves image super- resolution. In Proceedings of the Computer Vision and Pat- tern Recognition Conference, pages 23141–23152, 2025. 2

    2025

  41. [49]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models

    Chong Mou, Xintao Wang, Liangbin Xie, Yanze Wu, Jian Zhang, Zhongang Qi, and Ying Shan. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4296–4304, 2024. 2

    2024

  42. [50]

    Efficient real- 10 world image super-resolution via adaptive directional gradi- ent convolution

    Long Peng, Yang Cao, Renjing Pei, Wenbo Li, Jiaming Guo, Xueyang Fu, Yang Wang, and Zheng-Jun Zha. Efficient real- 10 world image super-resolution via adaptive directional gradi- ent convolution. arXiv preprint arXiv:2405.07023, 2024. 1

    2024 arXiv

  43. [51]

    Towards real- istic data generation for real-world super-resolution

    Long Peng, Wenbo Li, Renjing Pei, Jingjing Ren, Jiaqi Xu, Yang Wang, Yang Cao, and Zheng-Jun Zha. Towards real- istic data generation for real-world super-resolution. arXiv preprint arXiv:2406.07255, 2024. 2

    2024 arXiv

  44. [52]

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

    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 European Con- ference on Computer Vision, pages 285–303. Springer, 2024. 2

    2024

  45. [53]

    High-resolution image synthesis with latent diffusion models

    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. 1, 3, 5, 6

    2022

  46. [54]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. Advances in Neural Information...

    2022

  47. [55]

    Dcdm: Diffusion-conditioned-diffusion model for scene text image super-resolution

    Shrey Singh, Prateek Keserwani, Masakazu Iwamura, and Partha Pratim Roy. Dcdm: Diffusion-conditioned-diffusion model for scene text image super-resolution. In European Conference on Computer Vision , pages 303–320. Springer,

  48. [56]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. arXiv preprint arXiv:2011.13456, 2020. 2

    2011 arXiv

  49. [57]

    Stability.ai. Sd. https://stability.ai/stable- diffusion, 2021. 2, 4

    2021

  50. [58]

    Blindly assess image qual- ity in the wild guided by a self-adaptive hyper network

    Shaolin Su, Qingsen Yan, Yu Zhu, Cheng Zhang, Xin Ge, Jinqiu Sun, and Yanning Zhang. Blindly assess image qual- ity in the wild guided by a self-adaptive hyper network. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 3667–3676, 2020. 5

    2020

  51. [59]

    Improving the stability of dif- fusion models for content consistent super-resolution

    Lingchen Sun, Rongyuan Wu, Zhengqiang Zhang, Hong- wei Yong, and Lei Zhang. Improving the stability of dif- fusion models for content consistent super-resolution. arXiv preprint arXiv:2401.00877, 2023. 1

    2023 arXiv

  52. [60]

    Perception-distortion balanced super- resolution: A multi-objective optimization perspective

    Lingchen Sun, Jie Liang, Shuaizheng Liu, Hongwei Yong, and Lei Zhang. Perception-distortion balanced super- resolution: A multi-objective optimization perspective. IEEE Transactions on Image Processing, 2024. 1, 2

    2024

  53. [61]

    Pixel-level and semantic-level adjustable super-resolution: A dual-lora approach

    Lingchen Sun, Rongyuan Wu, Zhiyuan Ma, Shuaizheng Liu, Qiaosi Yi, and Lei Zhang. Pixel-level and semantic-level adjustable super-resolution: A dual-lora approach. arXiv preprint arXiv:2412.03017, 2024. 2

    2024 arXiv

  54. [62]

    Ex- ploring clip for assessing the look and feel of images

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Ex- ploring clip for assessing the look and feel of images. InPro- ceedings of the AAAI Conference on Artificial Intelligence , pages 2555–2563, 2023. 5

    2023

  55. [63]

    Exploiting diffusion prior for real-world image super-resolution

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution. arXiv preprint arXiv:2305.07015, 2023. 1, 2, 5

    2023 arXiv

  56. [64]

    Scene text image super-resolution in the wild

    Wenjia Wang, Enze Xie, Xuebo Liu, Wenhai Wang, Ding Liang, Chunhua Shen, and Xiang Bai. Scene text image super-resolution in the wild. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part X 16, pages 650–666. Springer,

    2020

  57. [65]

    Esrgan: En- hanced super-resolution generative adversarial networks

    Xintao Wang, Ke Yu, Shixiang Wu, Jinjin Gu, Yihao Liu, Chao Dong, Yu Qiao, and Chen Change Loy. Esrgan: En- hanced super-resolution generative adversarial networks. In Proceedings of the European conference on computer vision (ECCV) workshops, pages 0–0, 2018. 1, 2

    2018

  58. [66]

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

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 1905–1914,

    1905

  59. [67]

    Zero-shot im- age restoration using denoising diffusion null-space model

    Yinhuai Wang, Jiwen Yu, and Jian Zhang. Zero-shot im- age restoration using denoising diffusion null-space model. arXiv preprint arXiv:2212.00490, 2022. 2

    2022 arXiv

  60. [68]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 1, 5

    2004

  61. [69]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 2

    2004

  62. [70]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion. Advances in Neural Information Processing Systems , 36, 2024. 3

    2024

  63. [71]

    Perceive, understand and restore: Real-world image super- resolution with autoregressive multimodal generative mod- els

    Hongyang Wei, Shuaizheng Liu, Chun Yuan, and Lei Zhang. Perceive, understand and restore: Real-world image super- resolution with autoregressive multimodal generative mod- els. arXiv preprint arXiv:2503.11073, 2025. 1

    2025 arXiv

  64. [72]

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

    Pengxu Wei, Ziwei Xie, Hannan Lu, Zongyuan Zhan, Qixi- ang Ye, Wangmeng Zuo, and Liang Lin. Component divide- and-conquer for real-world image super-resolution. In Com- puter Vision–ECCV 2020: 16th European Conference, Glas- gow, UK, August 23–28, 2020, Proceedings, Part VIII ...

    2020

  65. [73]

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

    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. 5

    2023 arXiv

  66. [74]

    Seesr: Towards semantics- aware real-world image super-resolution

    Rongyuan Wu, Tao Yang, Lingchen Sun, Zhengqiang Zhang, Shuai Li, and Lei Zhang. Seesr: Towards semantics- aware real-world image super-resolution. arXiv preprint arXiv:2311.16518, 2023. 1, 2, 5

    2023 arXiv

  67. [75]

    One-step effective diffusion network for real-world image super-resolution

    Rongyuan Wu, Lingchen Sun, Zhiyuan Ma, and Lei Zhang. One-step effective diffusion network for real-world image super-resolution. arXiv preprint arXiv:2406.08177, 2024. 1, 2, 3, 5, 6

    2024 arXiv

  68. [76]

    Insvie-1m: Effective instruction-based video 11 editing with elaborate dataset construction

    Yuhui Wu, Liyi Chen, Ruibin Li, Shihao Wang, Chenxi Xie, and Lei Zhang. Insvie-1m: Effective instruction-based video 11 editing with elaborate dataset construction. arXiv preprint arXiv:2503.20287, 2025. 2

    2025 arXiv

  69. [77]

    Dnaedit: Direct noise alignment for text-guided rectified flow editing

    Chenxi Xie, Minghan Li, Shuai Li, Yuhui Wu, Qiaosi Yi, and Lei Zhang. Dnaedit: Direct noise alignment for text-guided rectified flow editing. arXiv preprint arXiv:2506.01430 ,

    arXiv

  70. [78]

    Desra: Detect and delete the artifacts of gan-based real-world super-resolution models

    Liangbin Xie, Xintao Wang, Xiangyu Chen, Gen Li, Ying Shan, Jiantao Zhou, and Chao Dong. Desra: Detect and delete the artifacts of gan-based real-world super-resolution models. In ICML, 2023. 2

    2023

  71. [79]

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

    Tao Yang, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic image super- resolution and personalized stylization. arXiv preprint arXiv:2308.14469, 2023. 1, 2, 5, 6

    2023 arXiv

  72. [80]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Micha ¨el Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 6613–6623...

    2024

  73. [81]

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

    Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiang- tao 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. arXiv preprint arXiv:2401.13627, 2024. 1, 2

    2024 arXiv

  74. [82]

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

    Zongsheng Yue, Jianyi Wang, and Chen Change Loy. Resshift: Efficient diffusion model for image super-resolution by residual shifting. arXiv preprint arXiv:2307.12348, 2023. 2

    2023 arXiv

  75. [83]

    Degradation-guided one-step im- age super-resolution with diffusion priors

    Aiping Zhang, Zongsheng Yue, Renjing Pei, Wenqi Ren, and Xiaochun Cao. Degradation-guided one-step im- age super-resolution with diffusion priors. arXiv preprint arXiv:2409.17058, 2024. 2, 3, 5

    2024 arXiv

  76. [84]

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

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timo- fte. 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. 1, 2

    2021

  77. [85]

    Fsim: A feature similarity index for image quality assess- ment

    Lin Zhang, Lei Zhang, Xuanqin Mou, and David Zhang. Fsim: A feature similarity index for image quality assess- ment. IEEE transactions on Image Processing, 20(8):2378– 2386, 2011. 5

    2011

  78. [86]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836–3847, 2023. 2

    2023

  79. [87]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 5

    2018

  80. [88]

    Efficient long-range attention network for image super- resolution

    Xindong Zhang, Hui Zeng, Shi Guo, and Lei Zhang. Efficient long-range attention network for image super- resolution. In European Conference on Computer Vision , pages 649–667. Springer, 2022. 2

    2022

  81. [89]

    Image super-resolution using very deep residual channel attention networks

    Yulun Zhang, Kunpeng Li, Kai Li, Lichen Wang, Bineng Zhong, and Yun Fu. Image super-resolution using very deep residual channel attention networks. In Proceedings of the European conference on computer vision (ECCV), pages 286–301, 2018. 1

    2018

  82. [90]

    Residual dense network for image super-resolution

    Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, and Yun Fu. Residual dense network for image super-resolution. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2472–2481, 2018. 1, 2

    2018

  83. [91]

    Diffusion-based blind text image super-resolution

    Yuzhe Zhang, Jiawei Zhang, Hao Li, Zhouxia Wang, Luwei Hou, Dongqing Zou, and Liheng Bian. Diffusion-based blind text image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25827–25836, 2024. 5, 6 12

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.