Pith. sign in

REVIEW 4 major objections 5 minor 71 references

Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read Scene text super-resolution can restore readable text and high image quality at once by first reconstructing glyph structures, then using them to guide full-image enhancement.

desk verdict Two-stage text-first SR is a real idea with honest ablations; the new UZ-ST benchmark is the thing to pressure before trusting the SOTA numbers. read the letter →

arxiv 2510.21590 v3 pith:RGGEU7OH submitted 2025-10-24 cs.CV

Minxing Luo , Linlong Fan , Wang Qiushi , Ge Wu , Yiyan Luo , Yuhang Yu , Jinwei Chen , Yaxing Wang
show 2 more authors
Qingnan Fan Jian Yang
This is my paper
classification cs.CV
keywords scenetextsuper-resolutionglyphstructureguidancetext-firstimage-latertwo-stagediffusionUZ-STdatasetextremezoomOCRaccuracyimagequality
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 tries to establish that a two-stage 'text-first, image-later' super-resolution framework can break the usual trade-off between readable text and overall image quality. TIGER first reconstructs precise glyph structures from low-resolution text regions, then feeds the recovered text mask as guidance into a full-image enhancement stage. On the existing Real-CE benchmark and their new UZ-ST dataset—the first Chinese scene-text benchmark with extreme zoom up to ×14.29—the authors report state-of-the-art OCR accuracy and image-quality scores. A sympathetic reader would care because real-world scene text is often severely degraded, and current generative methods tend to turn small Chinese characters into gibberish while making the background look plausible.

What carries the argument

The load-bearing mechanism is the two-stage cascade with a glyph structure mask as the intermediate representation. Stage 1 (text restoration) is a latent diffusion model with two output branches—one for appearance, one for text mask—whose mask branch is supervised by a combined segmentation loss and refined through a two-phase training strategy that first learns from synthetic plus real data, then freezes the output heads and trains only on synthetic data to clean up noisy real-world masks. Stage 2 (image enhancement) is a ControlNet-like diffusion model conditioned on the low-resolution image and the restored text mask, trained with MSE, LPIPS, and Sobel edge losses. The paper also builds

What would settle it

Randomly sample UZ-ST pairs, warp the LR image to the HR frame using the paper's alignment, and have human annotators mark corresponding control points; if the average residual error exceeds a few pixels on a meaningful fraction of samples, the dataset's alignment guarantee—and thus the quantitative comparison—would need revision. A simpler check: retrain TIGER without UZ-ST and compare the OCR-A delta against the full version to isolate the dataset's contribution.

Watch

Extended reading notes

Core claim

The central claim is that text and non-text should not be restored by one shared generative prior. TIGER explicitly decouples glyph restoration from image enhancement: a diffusion-based text restoration stage recovers stroke geometry in text regions, and the resulting glyph structure mask is injected as conditional guidance into a ControlNet-style full-image enhancer. The authors assert this ordering—text structure first, whole-image enhancement second—allows the enhancer to harmonize text with background without destroying strokes. They also present UZ-ST, a real-world dataset collected at four focal lengths (14mm to 200mm) with ×14.29 maximum zoom, containing 5,036 LR-HR pairs and 49,675 a

Load-bearing premise

The UZ-ST dataset is accurately pixel-aligned across all extreme-zoom pairs, so that each low-resolution image and its high-resolution ground truth truly depict the same scene; if residual misalignment remains, every UZ-ST training and evaluation number is compromised.

Editorial extensions

If this is right

  • Super-resolution systems for text-heavy scenes can adopt an explicit structure-restoration step rather than relying on a single generative prior, yielding both higher OCR accuracy and better perceptual image quality.
  • The UZ-ST dataset provides a benchmark for extreme zoom (×14.29) and multi-line Chinese text, enabling evaluation of SR methods under real-world degradations far beyond existing ×4 datasets.
  • The two-phase training schedule offers a recipe for using noisy real-world masks together with clean synthetic masks to train a structure-restoring diffusion model without costly pixel-level manual segmentation of degraded text.
  • Ablations show the method remains effective even with random or null OCR text input, implying the text restoration stage learns glyph structure from the low-resolution image itself rather than depending on the semantic condition.

Reading between the lines

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

  • The 'text-first, image-later' ordering may transfer to other structure-critical restoration tasks—such as face or document enhancement—where a dedicated structure-recovery module's output is used as conditioning for a general enhancer.
  • The UZ-ST alignment pipeline's claim of accurate pixel alignment is the least supported link: an independent human-verified alignment sample would be needed to rule out misalignment inflating the reported gains.
  • The OCR-A improvements shrink dramatically at ×14.29 (16.0% vs 63.2% at ×2.35), so practical deployments should expect the method's advantage to fade at the most extreme zoom levels.
  • Because evaluation relies on a single OCR engine, readability may be under- or over-estimated; testing with multiple recognizers or human readers would give a more robust measure of the claimed readability gains.
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

4 major / 5 minor

Summary. The paper proposes TIGER, a two-stage scene-text image super-resolution framework built on a 'text-first, image-later' paradigm. Stage 1 uses an OCR-conditioned latent diffusion model to restore glyph structure from cropped text regions and produces a text mask; Stage 2 uses a ControlNet-like network conditioned on the LR image and the predicted text mask to perform full-image super-resolution. The paper also introduces UZ-ST, a Chinese scene-text dataset with extreme zoom factors up to ×14.29, collected with a multi-focal-length phone camera and aligned by a cascade coarse-to-fine homography pipeline. Experiments on Real-CE and UZ-ST report state-of-the-art results in both image fidelity and OCR accuracy, and ablations study the role of the text mask, the OCR condition, and the UZ-ST training data. The central claim is that the two-stage decoupling breaks the usual trade-off between text readability and overall image quality.

Significance. If validated, TIGER would be a practically useful architecture for real-world scene-text SR, and UZ-ST would fill a clear gap as the first extreme-zoom, Chinese, multi-line text SR benchmark. The two-stage idea is well motivated, and the ablations are informative. In particular, the null-text / random-text experiments (Tables 5 and 10) are a strong and welcome response to the natural circularity concern: OCR-A remains high even when the OCR condition is uninformative, showing the model is not simply reading out the answer. The paper also reports extensive comparisons and states that code and data will be released. However, the central quantitative claims rest on the UZ-ST benchmark and on a modified Real-CE benchmark, and both raise methodological concerns that need to be resolved before the SOTA claim can be accepted.

major comments (4)
  1. [§4, Dataset split] The UZ-ST split, as described, appears to be at the image-pair level rather than the scene level: 'Among them, we randomly select 470, 589, and 581 pairs for evaluation under each zoom mode.' Since the same 200 mm image is the GT for the 14 mm, 35 mm, and 85 mm pairs of the same physical scene, a random split per zoom mode can place the same scene (and the same GT image) in both training and test sets. This would leak the exact target into training and inflate all UZ-ST numbers, especially OCR-A and PSNR. The paper must either demonstrate a scene-disjoint split or re-run the experiments on a properly split benchmark.
  2. [§4, Figure 3b, Table 8] The alignment validation is not sufficient for the load it carries. Table 8 reports PSNR/MSE/SSIM/NCC/AKD, but these are global similarity measures and do not verify pixel-level alignment; AKD=241.44 still suggests substantial residual keypoint displacement. The supplementary comparison is run only on the 35 mm subset, not on the hardest 14 mm (×14.29) mode. Because Stage 2 is trained with pixel-level MSE/LPIPS losses against the 200 mm GT, residual misalignment—especially depth-dependent parallax that homography cannot model—would blur supervision and bias every UZ-ST metric. Please provide per-zoom-mode residual alignment statistics, a human-verified subset, or an independent alignment check before using UZ-ST as the basis for the SOTA claim.
  3. [§5.1, §11] The paper filters and reannotates Real-CE, using only 337 training and 188 testing pairs, but the main text says 'Evaluations are performed on the Real-CE Benchmark [31]' without qualification. This changed benchmark is not comparable to the original Real-CE test set used in prior work. The authors should clearly state in the main text that all comparisons use their filtered subset, report the filtering criteria, and preferably also report results on the original Real-CE test split so that readers can relate the numbers to prior publications.
  4. [Tables 2–3] All quantitative results are reported as single runs without error bars or significance tests. Some of the decisive margins are small (e.g., 0.41 dB PSNR on UZ-ST, 2.6 pp OCR-A on Real-CE), and OCR recognition is stochastic (PP-OCRv3 is applied to generated images, which may also be stochastic). Reporting mean ± std over at least a few runs, or a paired significance test for the main comparisons, would materially strengthen the 'state-of-the-art' claim.
minor comments (5)
  1. [§4, §10] Focal length is inconsistently reported: the main text and Table 7 use 85 mm, while Section 10 says 84 mm. Please harmonize.
  2. [§4] The dataset name is misspelled as 'UltraZoom-Sence Text' in the Section 4 heading; should be 'Scene'.
  3. [Eq. (8), Table 3] ΔOCR-A is used in Table 3 but is never defined. Please define it explicitly, and state whether the LR baseline is the original LR input or the aligned LR used in the pipeline.
  4. [Eq. (4)] The text around Eq. (4) says the network 'denoises the zL only at the specific timestep t', but the formula is a single-step residual update. Please clarify the inference schedule and the role of the timestep t.
  5. [§5.2.1] The sentence 'we integrate the outputs of MARCONet and DiffTSR with HAT-generated results to simulate real-world application scenarios' is important for fairness but appears only in the experimental section. Please state this in the protocol description and justify that the integration is not disadvantageous to those baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: the OCR-conditioned text-restoration stage is explicitly ablated with null/random text, and the self-citations are baselines/implementation references, not load-bearing premises.

full rationale

The paper's derivation chain is not circular. The only step that could look self-referential is Stage 1 conditioning on OCR text while evaluating with OCR-A. However, the paper directly ablates this: Table 5 and Table 10 show that with null text or 100% random text the method still achieves OCR-A of 40.3-40.4% on the 35mm subset, outperforming TADiSR (35.5%). Thus the OCR-A result is not forced by the conditioning string; the model must still generate glyph structure from the LR image. The OCR-A metric compares against manually annotated ground-truth transcripts using Levenshtein distance, not against the conditioning string, and no fitted parameter is renamed as a prediction. Stage 2 uses a ControlNet-style condition with MSE, LPIPS, and edge losses against HR ground truth; the target image is not used as an input at inference. Self-citations such as TADiSR [19] and TSD-SR [13] are used as baselines or for implementation details (tile-based inference, SAM-TS mask extraction), not to justify the central 'text-first, image-later' claim, so they are not load-bearing. The UZ-ST alignment procedures and the small 35mm-only alignment comparison are a benchmark-quality and correctness risk, not a circularity: the paper does not derive its predictions from the same alignment metrics it reports. Overall, the central claims are supported by independent ablations and external benchmarks (Real-CE), and no circular step is exhibited.

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

The paper's contribution is empirical; it does not introduce free parameters in a physical model. The listed items are standard ML hyperparameters and data/modeling assumptions. No new entities are introduced.

free parameters (4)
  • λ_td, λ_Seg, λ_Focal, λ_Dice = 1, 0.1, 20, 1
    Stage-1 loss weights, chosen by hand; they balance diffusion, segmentation, and dice terms, but the architecture is not shown to be sensitive to exact values.
  • λ_l2, λ_LPIPS, λ_edge = 1, 5, 100 (λ_edge set to 0 on real data)
    Stage-2 loss weights; manual tuning to balance reconstruction, perceptual, and edge losses.
  • diffusion timestep t = 150
    Used in the single-step ControlNet sampling (Eq. 4); fixed based on experience.
  • learning rates = 5e-5 (stage 1), 5e-6 (stage 2)
    Standard small learning rates for fine-tuning diffusion models.
assumptions (3)
  • domain assumption UZ-ST cascade alignment yields pixel-accurate LR-HR pairs.
    Section 4 and Figure 3b; all quantitative results on UZ-ST assume alignment is correct. The supplementary compares alignment methods but does not show human-verified pixel accuracy.
  • domain assumption OCR-A measured by PP-OCRv3 is a valid proxy for textual readability.
    Eq. 8 uses Levenshtein ratio; the metric depends on a specific OCR engine's performance, which may not correlate perfectly with human perception.
  • domain assumption Synthetic degradations (Real-ESRGAN) capture enough real-world degradation for the first-stage training to generalize.
    Section 5.1 and Sec. 11; the whole two-phase training relies on this transfer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance." pith.science (2026). https://pith.science/paper/RGGEU7OH

@misc{pith2026251021590,
  author       = {Pith},
  title        = {Pith review of: Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RGGEU7OH}},
  note         = {Machine review of arXiv:2510.21590}
}
read the original abstract

Current image super-resolution methods show strong performance on natural images but distort text, creating a fundamental trade-off between image quality and textual readability. To address this, we introduce TIGER (Text-Image Guided supEr-Resolution), a novel two-stage framework that breaks this trade-off through a "text-first, image-later" paradigm. TIGER explicitly decouples glyph restoration from image enhancement: it first reconstructs precise text structures and uses them to guide full-image super-resolution. This ensures high fidelity and readability. To support comprehensive training and evaluation, we present the UZ-ST (UltraZoom-Scene Text) dataset, the first Chinese scene text dataset with extreme zoom. Extensive experiments show TIGER achieves state-of-the-art performance, enhancing readability and image quality.

Figures

Figures reproduced from arXiv: 2510.21590 by the authors.

Figure 1
Figure 1. We present TIGER (Text–Image Guided supEr-Resolution), a novel framework for scene text super-resolution. Its ‘text-first, image-later’ paradigm ensures accurate glyph restoration and consistently high overall image fidelity and visual quality. Abstract Current image super-resolution methods show strong performance on natural images but distort text, creating a fundamental trade-off between image quality and tex￾tua… view at source ↗
Figure 2
Figure 2. The framework of TIGER, which includes the Text Restoration stage (stage 1) and the Image Enhancement stage (stage 2). Stage [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of UZ-ST (UltraZoom-Scene Text). (a) Real-CE LRs show only mild degradation ( [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative Evaluation on Real-CE and UZ-ST. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative Results of Ablation Study with stage 2 fixed as the baseline. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Alignment method comparison. 3 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Additional qualitative results of TIGER. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Additional qualitative results of TIGER. [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Demonstration of our method under different languages. (Zoom in for more details.) [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Detailed examples of UZ-ST. 8 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 6 linked inside Pith

  1. [31]

    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. InICCV, pages 19452–19461, 2023. 2, 5, 6, 1

    2023

  2. [1]

    Dream- clear: High-capacity real-world image restoration with privacy-safe dataset curation.Advances in Neural Informa- tion Processing Systems, 37:55443–55469, 2024

    Yuang Ai, Xiaoqiang Zhou, Huaibo Huang, Xiaotian Han, Zhengyu Chen, Quanzeng You, and Hongxia Yang. Dream- clear: High-capacity real-world image restoration with privacy-safe dataset curation.Advances in Neural Informa- tion Processing Systems, 37:55443–55469, 2024. 2, 6

    2024

  3. [2]

    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. InProceedings of the IEEE/CVF international conference on computer vision, pages 3086–3095, 2019. 5, 2

    2019

  4. [3]

    Freeman, Michael Ru- binstein, Yuanzhen Li, and Dilip Krishnan

    Huiwen Chang, Han Zhang, Jarred Barber, Aaron Maschinot, Jos ´e Lezama, Lu Jiang, Ming-Hsuan Yang, Kevin Patrick Murphy, William T. Freeman, Michael Ru- binstein, Yuanzhen Li, and Dilip Krishnan. Muse: Text-to- image generation via masked generative transformers. InIn- ternational Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawai...

    2023

  5. [4]

    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. InCVPR, pages 12026–12035, 2021. 3

    2021

  6. [5]

    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. InCVPR, pages 22367–22377, 2023. 5

    2023

  7. [6]

    Effective diffusion transformer architecture for image super- resolution

    Kun Cheng, Lei Yu, Zhijun Tu, Xiao He, Liyu Chen, Yong Guo, Mingrui Zhu, Nannan Wang, Xinbo Gao, and Jie Hu. Effective diffusion transformer architecture for image super- resolution. InProceedings of the AAAI Conference on Arti- ficial Intelligence, pages 2455–2463, 2025. 2

    2025

  8. [7]

    Paddleocr 3.0 technical report, 2025

    Cheng Cui, Ting Sun, Manhui Lin, Tingquan Gao, Yubo Zhang, Jiaxuan Liu, Xueqing Wang, Zelun Zhang, Changda Zhou, Hongen Liu, Yue Zhang, Wenyu Lv, Kui Huang, Yichao Zhang, Jing Zhang, Jun Zhang, Yi Liu, Dianhai Yu, and Yanjun Ma. Paddleocr 3.0 technical report, 2025. 1

    2025

Show all 71 references
  1. [8]

    Textual alchemy: Coformer for scene text understanding

    Gayatri Deshmukh, Onkar Susladkar, Dhruv Makwana, Sparsh Mittal, et al. Textual alchemy: Coformer for scene text understanding. InProceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision, pages 2931–2941, 2024. 1

    2024

  2. [9]

    Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 1

    2021

  3. [10]

    Image quality assessment: Unifying structure and texture similarity.TPAMI, 44(5):2567–2581, 2020

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P Simoncelli. Image quality assessment: Unifying structure and texture similarity.TPAMI, 44(5):2567–2581, 2020. 5

    2020

  4. [11]

    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. InECCV, pages 184–199, 2014. 3

    2014

  5. [12]

    Boosting optical character recognition: A super- resolution approach.arXiv preprint, 2015

    Chao Dong, Ximei Zhu, Yubin Deng, Chen Change Loy, and Yu Qiao. Boosting optical character recognition: A super- resolution approach.arXiv preprint, 2015. 3

    2015

  6. [13]

    TSD-SR: one-step diffusion with target score distillation for real-world image super-resolution

    Linwei Dong, Qingnan Fan, Yihong Guo, Zhonghao Wang, Qi Zhang, Jinwei Chen, Yawei Luo, and Changqing Zou. TSD-SR: one-step diffusion with target score distillation for real-world image super-resolution. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 202...

    2025

  7. [14]

    Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution

    Linwei Dong, Qingnan Fan, Yihong Guo, Zhonghao Wang, Qi Zhang, Jinwei Chen, Yawei Luo, and Changqing Zou. Tsd-sr: One-step diffusion with target score distillation for real-world image super-resolution. InProceedings of the Computer Vision and Pattern Recognition Conference, p...

    2025

  8. [15]

    Dit4sr: Taming diffusion transformer for real-world image super-resolution

    Zheng-Peng Duan, Jiawei Zhang, Xin Jin, Ziheng Zhang, Zheng Xiong, Dongqing Zou, Jimmy Ren, Chun-Le Guo, and Chongyi Li. Dit4sr: Taming diffusion transformer for real-world image super-resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision,

  9. [16]

    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. InForty-first international conference on machi...

  10. [17]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.NeurIPS, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.NeurIPS, 30, 2017. 5

    2017

  11. [18]

    Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020. 1, 4

    2020

  12. [19]

    Text-aware real-world image super- resolution via diffusion model with joint segmentation de- coders.arXiv preprint arXiv:2506.04641, 2025

    Qiming Hu, Linlong Fan, Yiyan Luo, Yuhang Yu, Xiaojie Guo, and Qingnan Fan. Text-aware real-world image super- resolution via diffusion model with joint segmentation de- coders.arXiv preprint arXiv:2506.04641, 2025. 1, 3, 5, 6, 8

    2025 arXiv

  13. [20]

    Prestu: Pre-training for scene-text understanding

    Jihyung Kil, Soravit Changpinyo, Xi Chen, Hexiang Hu, Se- bastian Goodman, Wei-Lun Chao, and Radu Soricut. Prestu: Pre-training for scene-text understanding. InProceedings of the IEEE/CVF international conference on computer vision, pages 15270–15280, 2023. 1

    2023

  14. [21]

    Auto-encoding varia- tional bayes.arXiv preprint arXiv:1312.6114, 2013

    Diederik P Kingma and Max Welling. Auto-encoding varia- tional bayes.arXiv preprint arXiv:1312.6114, 2013. 3

    2013 arXiv

  15. [22]

    RIPE: Reinforcement Learning on Unlabeled Image Pairs for Ro- bust Keypoint Extraction.arXiv, 2025

    Johannes K ¨unzel, Anna Hilsmann, and Peter Eisert. RIPE: Reinforcement Learning on Unlabeled Image Pairs for Ro- bust Keypoint Extraction.arXiv, 2025. 5

    2025

  16. [23]

    Pp-ocrv3: More attempts for the improvement of ultra lightweight OCR sys- tem.CoRR, abs/2206.03001, 2022

    Chenxia Li, Weiwei Liu, Ruoyu Guo, Xiaoting Yin, Kaitao Jiang, Yongkun Du, Yuning Du, Lingfeng Zhu, Baohua Lai, Xiaoguang Hu, Dianhai Yu, and Yanjun Ma. Pp-ocrv3: More attempts for the improvement of ultra lightweight OCR sys- tem.CoRR, abs/2206.03001, 2022. 1

    2022 arXiv

  17. [24]

    Navigation-guided sparse scene representation for end-to-end autonomous driving

    Peidong Li and Dixiao Cui. Navigation-guided sparse scene representation for end-to-end autonomous driving. InThe Thirteenth International Conference on Learning Represen- tations, ICLR 2025, Singapore, April 24-28, 2025. OpenRe- view.net, 2025. 1

    2025

  18. [25]

    Learning generative structure prior for blind text image super-resolution

    Xiaoming Li, Wangmeng Zuo, and Chen Change Loy. Learning generative structure prior for blind text image super-resolution. InCVPR, pages 10103–10113, 2023. 1, 2, 3, 6 9

    2023

  19. [26]

    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. InCVPRW, pages 1775–1787, 2023. 5, 1

    2023

  20. [27]

    Diff- bir: Toward blind image restoration with generative diffusion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Bo Dai, Fanghua Yu, Yu Qiao, Wanli Ouyang, and Chao Dong. Diff- bir: Toward blind image restoration with generative diffusion prior. InECCV, pages 430–448, 2024. 1, 2, 6

    2024

  21. [28]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 1

    2017 arXiv

  22. [29]

    Object recognition from local scale-invariant features

    David G Lowe. Object recognition from local scale-invariant features. InProceedings of the seventh IEEE interna- tional conference on computer vision, pages 1150–1157. Ieee, 1999. 5

    1999

  23. [30]

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

    Jianqi Ma, Zhetong Liang, and Lei Zhang. A text atten- tion network for spatial deformation robust scene text image super-resolution. InCVPR, pages 5911–5920, 2022. 3

    2022

  24. [32]

    Plugnet: Degrada- tion aware scene text recognition supervised by a pluggable super-resolution unit

    Yongqiang Mou, Lei Tan, Hui Yang, Jingying Chen, Leyuan Liu, Rui Yan, and Yaohong Huang. Plugnet: Degrada- tion aware scene text recognition supervised by a pluggable super-resolution unit. InECCV, pages 158–174, 2020. 3

    2020

  25. [33]

    Hierarchical text-conditional image gen- eration with clip latents.arXiv preprint arXiv:2204.06125,

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gen- eration with clip latents.arXiv preprint arXiv:2204.06125,

    arXiv

  26. [34]

    Addison-Wesley Longman Publishing Co., Inc.,

    Richard A Roberts and Clifford T Mullis.Digital signal processing. Addison-Wesley Longman Publishing Co., Inc.,

  27. [35]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. InCVPR, pages 10684– 10695, 2022. 1, 3

    2022

  28. [36]

    Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

    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

  29. [37]

    Text-diae: a self-supervised degradation invariant autoencoder for text recognition and document enhancement

    Mohamed Ali Souibgui, Sanket Biswas, Andres Mafla, Ali Furkan Biten, Alicia Forn ´es, Yousri Kessentini, Josep Llad´os, Lluis Gomez, and Dimosthenis Karatzas. Text-diae: a self-supervised degradation invariant autoencoder for text recognition and document enhancement. Inprocee...

    2023

  30. [38]

    A scene-text synthesis engine achieved through learning from decomposed real-world data.IEEE Transactions on Image Processing, 32:5837–5851, 2023

    Zhengmi Tang, Tomo Miyazaki, and Shinichiro Omachi. A scene-text synthesis engine achieved through learning from decomposed real-world data.IEEE Transactions on Image Processing, 32:5837–5851, 2023. 5, 1

    2023

  31. [39]

    Anytext: Multilingual visual text genera- tion and editing

    Yuxiang Tuo, Wangmeng Xiang, Jun-Yan He, Yifeng Geng, and Xuansong Xie. Anytext: Multilingual visual text genera- tion and editing. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024. 1

    2024

  32. [40]

    Multi-task dif- fusion model for simultaneous text and image inpainting

    Bolin Wang, Xinyi Shen, and Kejun Zhang. Multi-task dif- fusion model for simultaneous text and image inpainting. In 2024 17th International Symposium on Computational Intel- ligence and Design (ISCID), pages 301–305. IEEE, 2024. 3

    2024

  33. [41]

    Exploiting diffusion prior for real-world image super-resolution.IJCV, 132(12):5929– 5949, 2024

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution.IJCV, 132(12):5929– 5949, 2024. 2

    2024

  34. [42]

    Textsr: Content-aware text super-resolution guided by recognition.arXiv preprint,

    Wenjia Wang, Enze Xie, Peize Sun, Wenhai Wang, Lixun Tian, Chunhua Shen, and Ping Luo. Textsr: Content-aware text super-resolution guided by recognition.arXiv preprint,

  35. [43]

    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. InECCV, pages 650–666, 2020. 2, 3

    2020

  36. [44]

    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. InICCV, pages 1905–1914, 2021. 2, 5, 1

    1905

  37. [45]

    Sinsr: diffusion-based image super- resolution in a single step

    Yufei Wang, Wenhan Yang, Xinyuan Chen, Yaohui Wang, Lanqing Guo, Lap-Pui Chau, Ziwei Liu, Yu Qiao, Alex C Kot, and Bihan Wen. Sinsr: diffusion-based image super- resolution in a single step. InCVPR, pages 25796–25805,

  38. [46]

    One-step effective diffusion network for real-world image super-resolution.NeurIPS, 37:92529–92553, 2024

    Rongyuan Wu, Lingchen Sun, Zhiyuan Ma, and Lei Zhang. One-step effective diffusion network for real-world image super-resolution.NeurIPS, 37:92529–92553, 2024. 1, 2, 6

    2024

  39. [47]

    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. InCVPR, pages 25456– 25467, 2024. 1, 2, 6

    2024

  40. [48]

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

    Tao Yang, Rongyuan Wu, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic im- age super-resolution and personalized stylization. InECCV, pages 74–91, 2024. 2

    2024

  41. [49]

    Hi-sam: Marrying segment anything model for hierarchical text segmentation

    Maoyuan Ye, Jing Zhang, Juhua Liu, Chenyu Liu, Baocai Yin, Cong Liu, Bo Du, and Dacheng Tao. Hi-sam: Marrying segment anything model for hierarchical text segmentation. TPAMI, 2024. 3, 4, 8, 1

    2024

  42. [50]

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

    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. InCVPR, pages 25669–25680, 2024. 1, 2, 5, 6

    2024

  43. [51]

    Benchmarking chinese text recognition: Datasets, baselines, and an empirical study.arXiv preprint arXiv:2112.15093, 2021

    Haiyang Yu, Jingye Chen, Bin Li, Jianqi Ma, Mengnan Guan, Xixi Xu, Xiaocong Wang, Shaobo Qu, and Xi- angyang Xue. Benchmarking chinese text recognition: Datasets, baselines, and an empirical study.arXiv preprint arXiv:2112.15093, 2021. 2

    2021 arXiv

  44. [52]

    Chinese text recognition with a pre-trained clip-like model through image-ids aligning

    Haiyang Yu, Xiaocong Wang, Bin Li, and Xiangyang Xue. Chinese text recognition with a pre-trained clip-like model through image-ids aligning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11943– 11952, 2023. 1

    2023

  45. [53]

    Resshift: Efficient diffusion model for image super- 10 resolution by residual shifting.NeurIPS, 36:13294–13307,

    Zongsheng Yue, Jianyi Wang, and Chen Change Loy. Resshift: Efficient diffusion model for image super- 10 resolution by residual shifting.NeurIPS, 36:13294–13307,

  46. [54]

    A normalized levenshtein distance metric.TPAMI, 29(6):1091–1095, 2007

    Li Yujian and Liu Bo. A normalized levenshtein distance metric.TPAMI, 29(6):1091–1095, 2007. 5

    2007

  47. [55]

    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. InICCV, pages 4791–4800, 2021. 2

    2021

  48. [56]

    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 ICCV, pages 3836–3847, 2023. 2, 4

    2023

  49. [57]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, pages 586–595,

  50. [58]

    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. InCVPR, pages 25827–25836,

  51. [59]

    Artbank: Artistic style transfer with pre-trained diffusion model and implicit style prompt bank

    Zhanjie Zhang, Quanwei Zhang, Wei Xing, Guangyuan Li, Lei Zhao, Jiakai Sun, Zehua Lan, Junsheng Luan, Yiling Huang, and Huaizhong Lin. Artbank: Artistic style transfer with pre-trained diffusion model and implicit style prompt bank. InProceedings of the AAAI conference on arti...

    2024

  52. [60]

    Reproducibility Statement To ensure reproducibility, we have made the following ef- forts:

  53. [61]

    We will release our code and dataset

  54. [62]

    5.1 and Sec

    We provide implementation details in Sec. 5.1 and Sec. 11, including the training process and selection of hyper-parameters

  55. [63]

    4 and Sec

    We provide details on evaluation metrics and dataset preparation in Sec. 4 and Sec. 10, and the code and data will be made available along with it

  56. [64]

    Ethics Statement This work focuses on improving scene text image super- resolution to support beneficial applications such as en- hancing accessibility, document restoration, and navigation assistance. However, we acknowledge potential risks, in- cluding misuse in privacy-sens...

  57. [65]

    All wording and factual content were reviewed and approved by the au- thors

    LLM Acknowledgments We thank ChatGPT (GPT-5) and DeepSeek for the help in refining the language and improving clarity. All wording and factual content were reviewed and approved by the au- thors

  58. [66]

    We first use PP-OCRV5 [7] for the rough annotation, then we manually filter the images and annotations

    Dataset Collection We use VIVO X200 Ultra to collect images for 4 separate focal lengths (14 mm, 35 mm, 84 mm, and 200 mm). We first use PP-OCRV5 [7] for the rough annotation, then we manually filter the images and annotations. During the fil- tering and annotation, each image...

  59. [67]

    Our synthetic dataset builds upon LSDIR [26], con- taining 27,000 triplets(x H , xL, xm)

    More Implementation Details Dataset Settings.For training, we combine a synthetic dataset with real paired datasets (Real-CE [31] and UZ- ST). Our synthetic dataset builds upon LSDIR [26], con- taining 27,000 triplets(x H , xL, xm). We render text on the GT of LSDIR and the co...

  60. [68]

    7, we provide detailed statistics on the composition of the UZ-ST dataset

    More details of UZ-ST In Tab. 7, we provide detailed statistics on the composition of the UZ-ST dataset. Additionally, in Fig. 10, we present 1 Table 7. Statistics of dataset size and line count in subsets of UltraZoom-ST. Subset image count line count mean lines/img img>5 lin...

  61. [69]

    During the comparison, we implement our strategy using SIFT and uti- lize the raw images from the 35mm dataset of our proposed UZ-ST dataset for evaluation

    Alignment Comparison We compare our proposed Cascade Coarse-to-Fine align- ment strategy with the alignment method proposed by Re- alSR [2] and single-time alignment using SIFT. During the comparison, we implement our strategy using SIFT and uti- lize the raw images from the 3...

  62. [70]

    Additional results We provide detailed quantitative results of UZ-ST in Tab. 11. Additional qualitative results are provided in Fig. 7 and Fig. 8. Fig. 9 also shows the ability of our method in handling other languages like Japanese and Korean. We also conduct additional ablat...

  63. [71]

    9, stage 1 is a standard diffusion process that takes multiple steps in inference, the efficiency of our model may be suboptimal compared to one-step methods

    Efficiency Analysis As shown in Tab. 9, stage 1 is a standard diffusion process that takes multiple steps in inference, the efficiency of our model may be suboptimal compared to one-step methods. However, our method achieves state-of-the-art performance in OCR-A, which cannot ...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.