Pith. sign in

REVIEW 3 major objections 6 minor 72 references

Quo Vadis Handwritten Text Generation for Handwritten Text Recognition?

T0 review · 3 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read Style-faithful synthetic handwriting beats diverse synthetic data when fewer than 130 real lines are available for fine-tuning.

desk verdict Useful benchmark, but the headline crossover claim (Emuru <130 images, DiffPen beyond) doesn't survive the paper's own tables. read the letter →

arxiv 2508.09936 v1 pith:W4KCNNVK submitted 2025-08-13 cs.CV cs.DL

classification cs.CVcs.DL
keywords HandwrittenTextRecognitionGenerationlow-resourcelearningsyntheticpretrainingdatastylefidelityhistoricalmanuscriptsdiffusionmodelsautoregressive
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

Handwritten Text Recognition models struggle on small historical collections whose handwriting differs from typical training data. This paper asks which style of synthetic handwriting generator—adversarial, diffusion, or autoregressive—produces the most useful pretraining images for such low-resource settings. It finds that when fewer than about 130 real transcribed lines are available, the autoregressive model Emuru, whose synthetic images most closely match the target handwriting style, gives the best recognition results after fine-tuning. With more real data, the more stylistically varied images from the diffusion model DiffPen become the better pretraining source, because diversity outweighs style similarity. The practical upshot is a quantitative guideline: match style for very small collections, prioritize variety once enough real lines exist.

What carries the argument

The central object is an evaluation pipeline rather than a single mathematical identity: each of three off-the-shelf styled HTG models—VATr++ (GAN), DiffPen (diffusion), Emuru (autoregressive)—generates line images conditioned on a few real style samples; these images pretrain the DefCRNN recognizer (a deformable-convolution CNN-LSTM line recognizer) before fine-tuning on fractions of real data. The load-bearing comparisons are the style-fidelity/readability scatter (HWD vs CER) and the fine-tuning curves across data fractions, which isolate the crossover at roughly 130 images.

What would settle it

Re-run the comparison with an autoregressive and a diffusion generator trained on the same synthetic corpus, both generating whole lines, and measure HTR CER after fine-tuning on fewer than 130 real images; if the diffusion model performs as well as or better than the autoregressive one, the paradigm-level conclusion fails. Alternatively, take DiffPen and filter its outputs to match Emuru's HWD distribution; if its low-data advantage does not disappear, style similarity is not the operative factor.

Watch

Extended reading notes

Core claim

Running a pretrain-then-fine-tune pipeline with the DefCRNN recognizer on three historical single-author datasets (Leopardi, Washington, Saint Gall), the paper reports a consistent crossover: Emuru, the autoregressive generator, yields the lowest CER when fine-tuning uses fewer than 130 real images, and DiffPen, the diffusion generator, yields the lowest CER beyond that threshold. The paper interprets this as style similarity driving performance in extreme low-data regimes, while synthetic-data diversity drives it once fine-tuning data is more plentiful. It also finds that filtering synthetic samples by handwriting distance (HWD) has no clear benefit, while readability-based filtering (CER)

Load-bearing premise

The three HTG models are treated as representatives of their generative paradigms, even though they differ in training-data composition, output granularity, and conditioning, so the observed ranking could be caused by those differences rather than by the paradigm.

Editorial extensions

If this is right

  • For collections with fewer than about 130 labeled lines, use a style-faithful autoregressive generator such as Emuru to build pretraining data; matching the target handwriting matters more than variety.
  • For collections with more labeled lines, a diffusion-based generator such as DiffPen gives better HTR fine-tuning results by providing stylistic diversity.
  • Filtering synthetic pretraining data by handwriting-style distance (HWD) does not improve recognition and can be skipped.
  • Readability-based filtering helps only when the threshold is relaxed (CER around 0.30); strict thresholds remove too many samples and hurt or prevent convergence.
  • In zero-shot direct transfer without fine-tuning, Emuru-generated data yields the best recognition, indicating style fidelity is the key factor when no real target labels are used.

Reading between the lines

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

  • An extension the paper leaves implicit: the paradigm-level ranking may be confounded by differences in training-data composition and output granularity among the three models, so the conclusion should be re-tested with generators matched on those axes.
  • The roughly 130-image crossover is likely tied to the recognizer and datasets used; a practical calibration curve could be built for a given archive by sweeping fine-tuning size before committing to a generator.
  • Emuru's strong style fidelity despite being trained only on synthetic data hints that diverse synthetic corpora may be a general route to zero-shot style generalization, not just a property of autoregressive architectures.
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.

Referee Report

3 major / 6 minor

Summary. This paper studies the use of synthetic handwritten text generated by three styled HTG models (VATr++, DiffPen, Emuru) as pretraining data for low-resource handwritten text recognition. The authors propose a pipeline that generates author-specific synthetic line images, filters them by readability (TrOCR CER) or style fidelity (HWD percentiles), pretrains a CRNN-based HTR model (DefCRNN), and then fine-tunes on progressively smaller subsets of three historical datasets (Leopardi, Washington, Saint Gall). The central claims are that Emuru yields the most style-faithful synthetic data and the best zero-shot and very-low-resource fine-tuning performance; that there is a threshold around 130 real fine-tuning images below which style similarity is crucial and above which diversity (and therefore DiffPen) becomes dominant; and that HWD-based filtering gives no clear benefit while moderate CER-based filtering is beneficial. The paper also reports generation metrics (FID, KID, HWD, DeltaCER) for the three generators.

Significance. If the empirical ranking and the 130-image threshold were reliable, the paper would provide a practically useful benchmark and actionable guidance for HTG-based pretraining for HTR, spanning multiple languages and fine-tuning regimes. The negative result on HWD filtering is also a useful data point, and the large experimental matrix is a strength. However, the central quantitative threshold is not consistently supported by the paper's own tables, and the absence of repeated runs plus a model-design confound prevent strong paradigm-level conclusions. The paper is best viewed as a valuable comparative study of three specific off-the-shelf systems, with conclusions that need to be rephrased and supported more carefully.

major comments (3)
  1. [Section 4.4, Tables 2–4] The claimed 'fewer than 130 real images' crossover is contradicted by the reported results. In Washington, taking the minimum CER over all filtering rows for each model, Emuru beats DiffPen at every fine-tuning size: 4.4 vs 5.1 at 50% (263 real images), 4.2 vs 4.6 at 75% (394), and 3.5 vs 3.9 at 100% (526). Thus Emuru wins even far above the stated 130-image threshold. In Leopardi, DiffPen only becomes better at 50% (652 images), not above 130: at 25% (326 images) Emuru's best CER is 10.0 vs DiffPen's 10.1. In Saint Gall the crossover occurs at 50% (234 images), not at 25% (117). These numbers also undermine the Section 5 recommendation that 'more than 130' training images makes DiffPen preferable. The 130-image cutoff appears to be an artifact of particular filtering choices and dataset-specific behavior, not a general empirical law.
  2. [Section 4.4, Tables 2–4] All CER numbers are single runs with no error bars or significance testing. Several load-bearing comparisons are small: e.g., 10.0 vs 10.1 on Leopardi at 25%, 4.2 vs 4.6 on Washington at 75%, and 3.5 vs 3.9 at 100%. Without repeated seeds, one cannot determine whether these differences are meaningful. The paper should report means and standard deviations over at least three seeds, or explicitly state that the close margins are not statistically assessed. This is particularly important because the threshold-based guideline in Section 4.4 and Section 5 depends on exactly these small gaps.
  3. [Section 3.2 and Section 5] The paper attributes performance differences to generative paradigms, but the three models differ in several other respects. VATr++ operates at word level with 15 style-sample inputs; DiffPen operates at word level and is patched into lines, uses 5 style samples, and was trained with a different data mixture; Emuru is a line-level autoregressive model trained exclusively on synthetic data. The observed advantage of Emuru in low-resource settings could stem from output granularity, training-data composition, or conditioning protocol rather than autoregression per se. Since the abstract and conclusions use paradigm labels ('adversarial', 'diffusion', 'autoregressive'), these confounds are load-bearing. The paper should either rephrase the conclusions to refer to 'the three compared systems' or provide additional controlled evidence that isolates the generative paradigm.
minor comments (6)
  1. [Section 4.4] The sentence claiming that Emuru-generated images 'exhibit the lowest readability according to the TrOCR model' is unclear and appears inconsistent with the stated DeltaCER values. If Emuru's images are harder for TrOCR despite having low absolute DeltaCER, the text should explain this explicitly; otherwise it is likely a typo for 'highest readability' or 'lowest CER gap.'
  2. [Section 4.4] The 130-image threshold is presented using approximate examples (10% of Leopardi corresponds to 130 images; 25% of Saint Gall corresponds to 125). Given its importance, the paper should specify exactly which fine-tuning fractions and datasets the threshold applies to, and how it was derived from the tables.
  3. [Section 3.3 / Section 5] The conclusion that HWD-based filtering gives no clear benefit should be qualified by the fact that HWD thresholds are percentile-based per model, not absolute style-fidelity cutoffs. Comparing the 'HWD25%' of one model with that of another does not compare equal absolute fidelity levels. An analysis with common absolute HWD thresholds, or a statement of this limitation, would strengthen the filtering section.
  4. [Figure 3] The caption reads 'the respected target dataset'; this should be 'the respective target dataset.'
  5. [References] Reference [50] has a truncated title: 'Alfie: Democratising RGBA Image Generation with No.' The full title should be restored.
  6. [General] The paper would benefit from a reproducibility section or link to code/checkpoints. The tables report the exact numbers, but no scripts or pretrained models are provided; for a benchmark-style study, releasing the pipeline would substantially increase the paper's value.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical benchmark whose ranking is measured by external HTR performance, not derived from fitted quantities or self-referential definitions.

full rationale

The paper's central claim—that Emuru provides the best pretraining data for HTR when fewer than ~130 real images are available, and that DiffPen becomes preferable with more data—is an empirical finding based on downstream CER of an external DefCRNN model, not a quantity fitted to the conclusion. The comparison uses fixed, pre-defined filtering thresholds (CER < 0.15/0.30/0.45/0.60 and HWD percentile cutoffs) and external metrics (FID, KID, TrOCR CER). The HTG models themselves are taken off-the-shelf from prior work; that several of those prior papers share authors is a self-evaluation concern but not a circular derivation, because the recognition results are obtained with an independently defined HTR model and standard benchmarks (Leopardi, Washington, Saint Gall). No load-bearing step reduces by construction to its inputs: style fidelity is measured by a published metric (HWD), readability by TrOCR, and the final ranking is read directly from tables of measured CER. The asserted 130-image crossover is not supported consistently by the paper's own tables, but that is an empirical-consistency/correctness issue, not circularity. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors, and no ansatz is smuggled in via self-citation as the basis of the main result.

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

The central claims rest on three hand-picked filtering thresholds, a post-hoc data-derived low-data threshold, and several domain assumptions about the validity of TrOCR CER and HWD as quality proxies. No new physical or model entities are introduced.

free parameters (3)
  • HWD percentile thresholds for style filtering = 25%, 50%, 75% (per model)
    Chosen as percentiles of each model's own HWD distribution to define strict/moderate/relaxed filtering; these thresholds are data-dependent and differ per generator.
  • CER filtering thresholds = 0.15, 0.30, 0.45, 0.60
    Hand-picked cutoffs on TrOCR CER to define readability filter settings.
  • Low-data threshold for style-fidelity benefit = 130 real images
    Post-hoc number inferred from the fine-tuning curves; the paper states that below this amount style similarity matters more. It is derived from the data it explains.
assumptions (4)
  • domain assumption TrOCR CER is a valid readability measure for generated historical handwriting.
    Used to define the readability filter and to interpret Emuru's low CER scores as style fidelity rather than illegibility.
  • domain assumption HWD captures style fidelity of generated handwriting.
    HWD is used both as an evaluation metric and as the basis for the style filtering; it was proposed by the authors in prior work.
  • domain assumption The three selected datasets are representative of low-resource single-author historical collections.
    The conclusions are drawn from three datasets spanning three languages and periods; no statistical framing treats them as a sample of a broader population.
  • domain assumption VATr++, DiffPen, and Emuru each represent their paradigm (GAN, diffusion, autoregressive) well enough to attribute performance differences to paradigm.
    The models differ in training data (Emuru is synthetic-only) and output granularity, so paradigm is confounded with model-specific design.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quo Vadis Handwritten Text Generation for Handwritten Text Recognition?." pith.science (2026). https://pith.science/paper/W4KCNNVK

@misc{pith2026250809936,
  author       = {Pith},
  title        = {Pith review of: Quo Vadis Handwritten Text Generation for Handwritten Text Recognition?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4KCNNVK}},
  note         = {Machine review of arXiv:2508.09936}
}
read the original abstract

The digitization of historical manuscripts presents significant challenges for Handwritten Text Recognition (HTR) systems, particularly when dealing with small, author-specific collections that diverge from the training data distributions. Handwritten Text Generation (HTG) techniques, which generate synthetic data tailored to specific handwriting styles, offer a promising solution to address these challenges. However, the effectiveness of various HTG models in enhancing HTR performance, especially in low-resource transcription settings, has not been thoroughly evaluated. In this work, we systematically compare three state-of-the-art styled HTG models (representing the generative adversarial, diffusion, and autoregressive paradigms for HTG) to assess their impact on HTR fine-tuning. We analyze how visual and linguistic characteristics of synthetic data influence fine-tuning outcomes and provide quantitative guidelines for selecting the most effective HTG model. The results of our analysis provide insights into the current capabilities of HTG methods and highlight key areas for further improvement in their application to low-resource HTR.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 70 canonical work pages

  1. [1]

    Boosting offline handwritten text recog- nition in historical documents with few labeled lines.IEEE Access, 2021

    José Carlos Aradillas, Juan José Murillo-Fuentes, and Pablo M Olmos. Boosting offline handwritten text recog- nition in historical documents with few labeled lines.IEEE Access, 2021. 1, 2

  2. [2]

    Handwrit- ing Recognition in Low-Resource Scripts Using Adversarial Learning

    Ayan Kumar Bhunia, Abhirup Das, Ankan Kumar Bhunia, Perla Sai Raj Kishore, and Partha Pratim Roy. Handwrit- ing Recognition in Low-Resource Scripts Using Adversarial Learning. InCVPR, 2019. 2

  3. [3]

    Handwriting Transformers

    Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Fahad Shahbaz Khan, and Mubarak Shah. Handwriting Transformers. InICCV, 2021. 1, 2

  4. [4]

    Sutherland, Michael Arbel, and Arthur Gretton

    Mikołaj Bi ´nkowski, Dougal J. Sutherland, Michael Arbel, and Arthur Gretton. Demystifying MMD GANs. InICLR,

  5. [5]

    Joint line segmentation and transcrip- tion for end-to-end handwritten paragraph recognition

    Théodore Bluche. Joint line segmentation and transcrip- tion for end-to-end handwritten paragraph recognition. In NeurIPS, 2016. 2

  6. [6]

    Gated convolu- tional recurrent neural networks for multilingual handwriting recognition

    Théodore Bluche and Ronaldo Messina. Gated convolu- tional recurrent neural networks for multilingual handwriting recognition. InICDAR, 2017. 2

  7. [7]

    Scan, Attend and Read: End-to-End Handwrit- ten Paragraph Recognition with MDLSTM Attention

    Théodore Bluche, Jérôome Louradour, and Ronaldo Messina. Scan, Attend and Read: End-to-End Handwrit- ten Paragraph Recognition with MDLSTM Attention. InIC- DAR, 2017. 2

  8. [8]

    Learning to Read L’Infinito: Handwritten Text Recognition with Synthetic Training Data

    Silvia Cascianelli, Marcella Cornia, Lorenzo Baraldi, Maria Ludovica Piazzi, Rosiana Schiuma, and Rita Cuc- chiara. Learning to Read L’Infinito: Handwritten Text Recognition with Synthetic Training Data. InCAIP, 2021. 1, 2, 3, 5

Show all 72 references
  1. [9]

    Boosting Modern and Historical Handwrit- ten Text Recognition with Deformable Convolutions.IJDAR, pages 1–15, 2022

    Silvia Cascianelli, Marcella Cornia, Lorenzo Baraldi, Rita Cucchiara, et al. Boosting Modern and Historical Handwrit- ten Text Recognition with Deformable Convolutions.IJDAR, pages 1–15, 2022. 3

  2. [10]

    The lam dataset: A novel benchmark for line- level handwritten text recognition

    Silvia Cascianelli, Vittorio Pippi, Maarand Martin, Marcella Cornia, Lorenzo Baraldi, Kermorvant Christopher, and Rita Cucchiara. The lam dataset: A novel benchmark for line- level handwritten text recognition. InICPR, 2022. 1

  3. [11]

    Handwriting recognition of historical documents with few labeled data

    Edgard Chammas, Chafic Mokbel, and Laurence Likforman- Sulem. Handwriting recognition of historical documents with few labeled data. 2018. 2

  4. [12]

    A ranking-based feature selection approach for handwritten character recogni- tion.Pattern Recognit

    Nicole Dalia Cilia, Claudio De Stefano, Francesco Fontanella, and Alessandra Scotto di Freca. A ranking-based feature selection approach for handwritten character recogni- tion.Pattern Recognit. Lett., pages 77–86, 2019. 2

  5. [13]

    KuroNet: Pre-Modern Japanese Kuzushiji Character Recog- nition with Deep Learning

    Tarin Clanuwat, Alex Lamb, and Asanobu Kitamoto. KuroNet: Pre-Modern Japanese Kuzushiji Character Recog- nition with Deep Learning. InICDAR, 2019. 2

  6. [14]

    Clark, Dan Garrette, Iulia Turc, and John Wi- eting

    Jonathan H. Clark, Dan Garrette, Iulia Turc, and John Wi- eting. CANINE: Pre-training an Efficient Tokenization-Free Encoder for Language Representation.Transactions of the Association for Computational Linguistics, 10:73–91, 2022. 4

  7. [15]

    Watch Your Strokes: Improving Handwritten Text Recognition with Deformable Convolutions

    Iulian Cojocaru, Silvia Cascianelli, Lorenzo Baraldi, Massi- miliano Corsini, and Rita Cucchiara. Watch Your Strokes: Improving Handwritten Text Recognition with Deformable Convolutions. InICPR, 2020. 3

  8. [16]

    Recurrence-free unconstrained handwritten text recognition using gated fully convolutional network

    Denis Coquenet, Clément Chatelain, and Thierry Paquet. Recurrence-free unconstrained handwritten text recognition using gated fully convolutional network. InICFHR, 2020. 2

  9. [17]

    Deformable convolutional networks

    Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. InCVPR, 2017. 3

  10. [18]

    Improving handwritten OCR with training samples generated by glyph conditional denoising diffusion probabilistic model

    Haisong Ding, Bozhi Luan, Dongnan Gui, Kai Chen, and Qiang Huo. Improving handwritten OCR with training samples generated by glyph conditional denoising diffusion probabilistic model. InICDAR, 2023. 2

  11. [19]

    Transcription alignment of Latin manuscripts using hidden Markov models

    Andreas Fischer, V olkmar Frinken, Alicia Fornés, and Horst Bunke. Transcription alignment of Latin manuscripts using hidden Markov models. InHIP, 2011. 5

  12. [20]

    Lexicon-free handwritten word spotting using character HMMs.Pattern Recognit

    Andreas Fischer, Andreas Keller, V olkmar Frinken, and Horst Bunke. Lexicon-free handwritten word spotting using character HMMs.Pattern Recognit. Lett., pages 934–942,

  13. [21]

    ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation

    Sharon Fogel, Hadar Averbuch-Elor, Sarel Cohen, Shai Ma- zor, and Roee Litman. ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation. InCVPR,

  14. [22]

    HiGAN: Handwriting Imita- tion Conditioned on Arbitrary-Length Texts and Disentan- gled Styles

    Ji Gan and Weiqiang Wang. HiGAN: Handwriting Imita- tion Conditioned on Arbitrary-Length Texts and Disentan- gled Styles. InAAAI, 2021. 2

  15. [23]

    Hi- GAN+: Handwriting Imitation GAN with Disentangled Rep- resentations.ACM Trans

    Ji Gan, Weiqiang Wang, Jiaxu Leng, and Xinbo Gao. Hi- GAN+: Handwriting Imitation GAN with Disentangled Rep- resentations.ACM Trans. Graphics, pages 1–17, 2022. 2

  16. [24]

    Generative Adversarial Nets

    Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C Courville, and Yoshua Bengio. Generative Adversarial Nets. In NeurIPS, 2014. 3

  17. [25]

    Transfer learning for handwriting recognition on historical documents

    Adeline Granet, Emmanuel Morin, Harold Mouchère, Solen Quiniou, and Christian Viard-Gaudin. Transfer learning for handwriting recognition on historical documents. In ICPRAM, 2018. 1, 2

  18. [26]

    Offline handwrit- ing recognition with multidimensional recurrent neural net- works

    Alex Graves and Jürgen Schmidhuber. Offline handwrit- ing recognition with multidimensional recurrent neural net- works. InNeurIPS, 2009. 2

  19. [27]

    GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

    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 Equilibrium. InNeurIPS, 2017. 5

  20. [28]

    Boosting handwriting text recognition in small databases with transfer learning

    José Carlos Aradillas Jaramillo, Juan José Murillo-Fuentes, and Pablo M Olmos. Boosting handwriting text recognition in small databases with transfer learning. InICFHR, 2018. 2

  21. [29]

    GANwriting: Content- Conditioned Generation of Styled Handwritten Word Im- ages

    Lei Kang, Pau Riba, Yaxing Wang, Marçal Rusiñol, Ali- cia Fornés, and Mauricio Villegas. GANwriting: Content- Conditioned Generation of Styled Handwritten Word Im- ages. InECCV, 2020. 2

  22. [30]

    Content and style aware generation of text-line images for handwriting recognition.IEEE Trans

    Lei Kang, Pau Riba, Marcal Rusinol, Alicia Fornes, and Mauricio Villegas. Content and style aware generation of text-line images for handwriting recognition.IEEE Trans. PAMI, pages 1–1, 2021. 2

  23. [31]

    Pay attention to what you read: non- recurrent handwritten text-line recognition.Pattern Recog- nit., 129:108766, 2022

    Lei Kang, Pau Riba, Marçal Rusiñol, Alicia Fornés, and Mauricio Villegas. Pay attention to what you read: non- recurrent handwritten text-line recognition.Pattern Recog- nit., 129:108766, 2022. 1, 2

  24. [32]

    Auto-Encoding Vari- ational Bayes

    Diederik P Kingma and Max Welling. Auto-Encoding Vari- ational Bayes. InICLR, 2013. 4

  25. [33]

    TextStyleBrush: Transfer of Text Aesthetics from a Single Example.IEEE Trans

    Praveen Krishnan, Rama Kovvuri, Guan Pang, Boris Vas- silev, and Tal Hassner. TextStyleBrush: Transfer of Text Aesthetics from a Single Example.IEEE Trans. PAMI, 2023. 2

  26. [34]

    TrOCR: Transformer-based optical character recognition with pre- trained models.AAAI, 2023

    Minghao Li, Tengchao Lv, Lei Cui, Yijuan Lu, Dinei Flo- rencio, Cha Zhang, Zhoujun Li, and Furu Wei. TrOCR: Transformer-based optical character recognition with pre- trained models.AAAI, 2023. 1, 2, 4, 5

  27. [35]

    Diffusion Models for Hand- writing Generation.arXiv preprint arXiv:2011.06704, 2020

    Troy Luhman and Eric Luhman. Diffusion Models for Hand- writing Generation.arXiv preprint arXiv:2011.06704, 2020. 2

  28. [36]

    Conditional Generative Adversarial Nets.arXiv preprint arXiv:1411.1784, 2014

    Mehdi Mirza and Simon Osindero. Conditional Generative Adversarial Nets.arXiv preprint arXiv:1411.1784, 2014. 3

  29. [37]

    Full-page text recognition: Learning where to start and when to stop

    Bastien Moysset, Christopher Kermorvant, and Christian Wolf. Full-page text recognition: Learning where to start and when to stop. InICDAR, 2017. 2

  30. [38]

    A Survey of Historical Document Im- age Datasets.IJDAR, 25(4):305–338, 2022

    Konstantina Nikolaidou, Mathias Seuret, Hamam Mokayed, and Marcus Liwicki. A Survey of Historical Document Im- age Datasets.IJDAR, 25(4):305–338, 2022. 1

  31. [39]

    WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models

    Konstantina Nikolaidou, George Retsinas, Vincent Christlein, Mathias Seuret, Giorgos Sfikas, Elisa Barney Smith, Hamam Mokayed, and Marcus Liwicki. WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models. InICDAR, 2023. 2

  32. [40]

    DiffusionPen: Towards Controlling the Style of Handwritten Text Generation.ECCV, 2024

    Konstantina Nikolaidou, George Retsinas, Giorgos Sfikas, and Marcus Liwicki. DiffusionPen: Towards Controlling the Style of Handwritten Text Generation.ECCV, 2024. 1, 3, 4

  33. [41]

    Rethinking HTG Evaluation: Bridg- ing Generation and Recognition.ECCVW, 2024

    Konstantina Nikolaidou, George Retsinas, Giorgos Sfikas, and Marcus Liwicki. Rethinking HTG Evaluation: Bridg- ing Generation and Recognition.ECCVW, 2024. 1, 4, 7

  34. [42]

    Dropout improves recurrent neural net- works for handwriting recognition

    Vu Pham, Théodore Bluche, Christopher Kermorvant, and Jérôme Louradour. Dropout improves recurrent neural net- works for handwriting recognition. InICFHR, 2014. 2

  35. [43]

    Evaluating Synthetic Pre-Training for Handwrit- ing Processing Tasks.Pattern Recognit

    Vittorio Pippi, Silvia Cascianelli, Lorenzo Baraldi, and Rita Cucchiara. Evaluating Synthetic Pre-Training for Handwrit- ing Processing Tasks.Pattern Recognit. Lett., 2023. 1

  36. [44]

    Hand- written Text Generation from Visual Archetypes

    Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Hand- written Text Generation from Visual Archetypes. InCVPR,

  37. [45]

    How to choose pretrained handwriting recognition models for single writer fine-tuning

    Vittorio Pippi, Silvia Cascianelli, Christopher Kermorvant, and Rita Cucchiara. How to choose pretrained handwriting recognition models for single writer fine-tuning. InICDAR,

  38. [46]

    HWD: A Novel Evaluation Score for Styled Handwritten Text Generation

    Vittorio Pippi, Fabio Quattrini, Silvia Cascianelli, and Rita Cucchiara. HWD: A Novel Evaluation Score for Styled Handwritten Text Generation. InBMVC, 2023. 3, 4, 5

  39. [47]

    Zero-Shot Styled Text Image Generation, but Make It Autoregressive

    Vittorio Pippi, Fabio Quattrini, Silvia Cascianelli, Alessio Tonioni, and Rita Cucchiara. Zero-Shot Styled Text Image Generation, but Make It Autoregressive. InCVPR, 2025. 1, 2, 3, 4

  40. [48]

    Are multidimensional recurrent layers re- ally necessary for handwritten text recognition? InICDAR,

    Joan Puigcerver. Are multidimensional recurrent layers re- ally necessary for handwritten text recognition? InICDAR,

  41. [49]

    V olumetric fast fourier convolution for detecting ink on the carbonized herculaneum papyri

    Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. V olumetric fast fourier convolution for detecting ink on the carbonized herculaneum papyri. InICCVW, pages 1726–1734, 2023. 1

  42. [50]

    Alfie: Democratising RGBA Image Generation with No

    Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Alfie: Democratising RGBA Image Generation with No. InECCVW, pages 38–55. Springer, 2024. 4

  43. [51]

    Binarizing documents by leveraging both space and frequency

    Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Binarizing documents by leveraging both space and frequency. InICDAR, pages 3–22. Springer, 2024. 1

  44. [52]

    Merging and splitting diffusion paths for se- mantically coherent panoramas

    Fabio Quattrini, Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Merging and splitting diffusion paths for se- mantically coherent panoramas. InECCV, pages 234–251. Springer, 2024. 4

  45. [53]

    In ECCVW, pages 212–228, 2024

    Fabio Quattrini, Carmine Zaccagnino, Silvia Cascianelli, Laura Righi, and Rita Cucchiara.µgat: Improving Single- Page Document Parsing by Providing Multi-page Context. In ECCVW, pages 212–228, 2024. 1

  46. [54]

    From Seq2Seq Recognition to Handwritten Word Embeddings

    George Retsinas, Giorgos Sfikas, Christophoros Nikou, and Petros Maragos. From Seq2Seq Recognition to Handwritten Word Embeddings. InBMVC, 2021. 1

  47. [55]

    Best Practices for a Handwritten Text Recognition System

    George Retsinas, Giorgos Sfikas, Basilis Gatos, and Christophoros Nikou. Best Practices for a Handwritten Text Recognition System. InInternational Workshop on Docu- ment Analysis Systems, pages 247–259. Springer, 2022

  48. [56]

    Enhancing CRNN HTR Architectures with Trans- former Blocks

    George Retsinas, Konstantina Nikolaidou, and Giorgos Sfikas. Enhancing CRNN HTR Architectures with Trans- former Blocks. InICDAR, pages 425–440. Springer, 2024. 1

  49. [57]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image syn- thesis with latent diffusion models. InCVPR, 2022. 4

  50. [58]

    U- Net: Convolutional Networks for Biomedical Image Seg- mentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- Net: Convolutional Networks for Biomedical Image Seg- mentation. InMICCAI, pages 234–241. Springer, 2015. 4

  51. [59]

    A method of synthesiz- ing handwritten chinese images for data augmentation

    Xi Shen and Ronaldo Messina. A method of synthesiz- ing handwritten chinese images for data augmentation. In ICFHR, 2016. 2

  52. [60]

    An end-to-end trainable neural network for image-based sequence recogni- tion and its application to scene text recognition.IEEE Trans

    Baoguang Shi, Xiang Bai, and Cong Yao. An end-to-end trainable neural network for image-based sequence recogni- tion and its application to scene text recognition.IEEE Trans. PAMI, pages 2298–2304, 2016. 2, 3

  53. [61]

    One-shot Compositional Data Generation for Low Resource Handwritten Text Recogni- tion

    Mohamed Ali Souibgui, Ali Furkan Biten, Sounak Dey, Ali- cia Fornes, Yousri Kessentini, Lluis Gomez, Dimosthenis Karatzas, and Josep Llados. One-shot Compositional Data Generation for Low Resource Handwritten Text Recogni- tion. InWACV, 2022. 2

  54. [62]

    Improving text recognition using optical and language model writer adaptation

    Yann Soullard, Wassim Swaileh, Pierrick Tranouez, Thierry Paquet, and Clement Chatelain. Improving text recognition using optical and language model writer adaptation. InIC- DAR, 2019. 2

  55. [63]

    Fully convolutional networks for handwriting recognition

    Felipe Petroski Such, Dheeraj Peri, Frank Brockler, Hutkowski Paul, and Raymond Ptucha. Fully convolutional networks for handwriting recognition. InICFHR, 2018. 2

  56. [64]

    V ATr++: Choose Your Words Wisely for Handwritten Text Genera- tion.IEEE Trans

    Bram Vanherle, Vittorio Pippi, Silvia Cascianelli, Nick Michiels, Frank Van Reeth, and Rita Cucchiara. V ATr++: Choose Your Words Wisely for Handwritten Text Genera- tion.IEEE Trans. PAMI, 2024. 1, 2, 3

  57. [65]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. InNeurIPS, 2017. 2

  58. [66]

    Handwriting recognition with large multidimensional long short-term memory recurrent neural networks

    Paul V oigtlaender, Patrick Doetsch, and Hermann Ney. Handwriting recognition with large multidimensional long short-term memory recurrent neural networks. InICFHR,

  59. [67]

    Rescoring Sequence-to-Sequence Models for Text Line Recognition with CTC-Prefixes.arXiv preprint arXiv:2110.05909, 2021

    Christoph Wick, Jochen Zöllner, and Tobias Grün- ing. Rescoring Sequence-to-Sequence Models for Text Line Recognition with CTC-Prefixes.arXiv preprint arXiv:2110.05909, 2021. 1

  60. [68]

    Trans- former for Handwritten Text Recognition Using Bidirec- tional Post-decoding

    Christoph Wick, Jochen Zöllner, and Tobias Grüning. Trans- former for Handwritten Text Recognition Using Bidirec- tional Post-decoding. InICDAR, 2021. 2

  61. [69]

    Data augmentation for recog- nition of handwritten words and lines using a CNN-LSTM network

    Curtis Wigington, Seth Stewart, Brian Davis, Bill Barrett, Brian Price, and Scott Cohen. Data augmentation for recog- nition of handwritten words and lines using a CNN-LSTM network. InICDAR, 2017. 2

  62. [70]

    Start, Follow, Read: End-to-End Full-Page Handwriting Recognition

    Curtis Wigington, Chris Tensmeyer, Brian Davis, William Barrett, Brian Price, and Scott Cohen. Start, Follow, Read: End-to-End Full-Page Handwriting Recognition. InECCV,

  63. [71]

    OrigamiNet: Weakly- Supervised, Segmentation-Free, One-Step, Full Page Text Recognition by learning to unfold

    Mohamed Yousef and Tom E Bishop. OrigamiNet: Weakly- Supervised, Segmentation-Free, One-Step, Full Page Text Recognition by learning to unfold. InCVPR, 2020. 2

  64. [72]

    Conditional Text Image Generation with Diffu- sion Models

    Yuanzhi Zhu, Zhaohai Li, Tianwei Wang, Mengchao He, and Cong Yao. Conditional Text Image Generation with Diffu- sion Models. InCVPR, 2023. 2

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.