Pith. sign in

REVIEW 3 major objections 3 minor 1 cited by

Dual Orthogonal Guidance for Robust Diffusion-based Handwritten Text Generation

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

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

Pith's one-line read This paper proposes Dual Orthogonal Guidance (DOG), a sampling-time method that uses an orthogonal projection between positive and negatively perturbed prompts to reduce artifacts and increase style diversity in diffusion-based handwritten

desk verdict A plausible new guidance mechanism for handwriting generation that currently lacks visible proof of its central geometric claim; worth a referee's time. read the letter →

arxiv 2508.17017 v1 pith:B6BPTDMI submitted 2025-08-23 cs.CV

Konstantina Nikolaidou , George Retsinas , Giorgos Sfikas , Silvia Cascianelli , Rita Cucchiara , Marcus Liwicki This is my paper
classification cs.CV
keywords handwrittentextgenerationdiffusionmodelsclassifier-freeguidancesamplingorthogonalprojectionstylevariabilityartifactreductiontext-conditionedimagesynthesis
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 fix a known failure mode in diffusion-based handwritten text generation: models memorize training samples and produce distorted, unreadable letters, especially for uncommon words and challenging styles. It proposes a sampling guidance strategy called Dual Orthogonal Guidance (DOG), which combines a positive prompt with the orthogonal projection of a negatively perturbed prompt onto it. That projection is meant to define a direction in the model's conditioning space that repels artifacts without pulling generation away from the intended content. A triangular schedule weakens the guidance at the start and end of denoising, where the process is most fragile, and strengthens it in the middle. If DOG works as claimed, it improves content clarity and style variability without retraining, on top of existing diffusion models.

What carries the argument

Dual Orthogonal Guidance (DOG) is the load-bearing mechanism: a conditioning direction computed as the orthogonal projection of a negatively perturbed prompt embedding onto the original positive prompt embedding, then subtracted from the positive prompt to push the sample away from artifact-prone regions of latent space. The triangular schedule over the denoising trajectory controls when that push is applied, with weak guidance at the first and last steps and strong guidance in the middle, where the process is most sensitive.

What would settle it

Replace the orthogonal projection direction with a random vector of the same norm and rerun the experiments: if OCR accuracy and readability scores do not worsen, the orthogonality is not doing causal work. Alternatively, check the actual angle between the projected negative prompt and the positive prompt in the embedding space; if it is not close to 90 degrees, the 'orthogonal' label is misleading and the mechanism would need to be restated.

Watch

Extended reading notes

Core claim

The central claim is that replacing the standard classifier-free guidance direction with a dual orthogonal guidance direction yields a more stable, disentangled control signal for handwritten text generation. Specifically, the method takes the original text-and-style prompt (positive), constructs a negatively perturbed prompt by weakening or altering the conditioning, projects that negative prompt's embedding onto the positive prompt's embedding, and subtracts the projection to obtain an artifact-repelling direction. Applied via a triangular schedule over denoising steps, this direction steers generation away from distortions and memorized samples while preserving legibility, and increases s

Load-bearing premise

The orthogonal projection of a negatively perturbed prompt onto the positive prompt produces a semantically meaningful direction that repels artifacts, rather than adding noise or pulling the sample off-content.

Editorial extensions

If this is right

  • Existing diffusion-based handwriting models can be improved at inference time by swapping classifier-free guidance for DOG, with no retraining.
  • Generated handwritten words should become more readable to OCR systems and human readers, especially for rare or out-of-vocabulary words.
  • Style variation within the same content should increase without sacrificing legibility, because the orthogonal direction pushes away from memorized samples.
  • The triangular schedule suggests a general principle for diffusion guidance: minimize the guidance signal when the latent is most fragile, early and late in denoising.

Reading between the lines

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

  • The orthogonality assumption is geometric; it implicitly treats the diffusion conditioning space as Euclidean enough that subtracting the projection really removes the negative-prompt component. A natural test is to measure the angle between the projected negative prompt and the positive prompt in the actual embedding space and correlate it with artifact rejection.
  • DOG could transfer beyond handwriting: the recipe of 'repel artifacts by orthogonalizing a negative prompt' may reduce hallucinations or distortions in general text-to-image generation.
  • The triangular schedule implies that the middle denoising steps are where high-level layout is decided; if so, tuning the peak position of the schedule could yield larger gains than tuning overall guidance strength.
  • An implicit limitation worth probing is the severity of the negative perturbation: if the perturbed prompt is too corrupted, its projection could degenerate and add noise, so robustness to that severity is a natural next experiment.
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 / 3 minor

Summary. The paper proposes Dual Orthogonal Guidance (DOG), a sampling guidance strategy for diffusion-based handwritten text generation. The method uses an orthogonal projection of a negatively perturbed prompt onto the original positive prompt to obtain a 'disentangled' guidance direction that steers generation away from artifacts while preserving content, and applies a triangular schedule to vary guidance strength across denoising steps. The authors claim that DOG improves content clarity and style variability on the DiffusionPen and One-DM benchmarks, including for out-of-vocabulary words and challenging styles, and that it is more stable than Classifier-Free Guidance at high guidance scales.

Significance. If the claims hold, DOG would provide a practical alternative to CFG for handwritten text generation, potentially improving readability and style diversity for rare words and difficult styles—areas where current diffusion HTG methods reportedly underperform. However, the manuscript as provided contains only the abstract, with no method details, derivations, ablations, or quantitative results. The central geometric claim about orthogonal projection is both novel and non-obvious, but its validity cannot be assessed from the visible evidence. The paper's significance therefore remains purely potential until the full method and experiments are supplied.

major comments (3)
  1. [Abstract] The core mechanism—'orthogonal projection of a negatively perturbed prompt onto the original positive prompt'—is undefined. Which embedding space is used? What inner product defines orthogonality? How are the 'positive prompt' and 'negatively perturbed prompt' represented and combined? No mathematical characterization or intuitive justification is given for why the projected direction should be orthogonal to content variation and aligned with artifact variation. Without this specification, the 'disentangled direction' claim is not testable. This is load-bearing because the entire contribution rests on this geometric assumption.
  2. [Abstract] The experimental claim that DOG 'improves both content clarity and style variability, even for out-of-vocabulary words and challenging writing styles' is stated without any supporting numbers. There are no effect sizes, error bars, baseline comparisons, or ablations isolating DOG from the triangular schedule or from the CFG baseline. The abstract asserts that DOG outperforms CFG, but no evidence is visible. The claim is falsifiable in principle, but as written it is unsupported.
  3. [Abstract] The abstract states that CFG 'produces noise at high guidance scales.' This is presented as a known limitation, but no reference or experiment is provided. More importantly, the abstract does not explain how DOG avoids this issue: is the triangular schedule responsible, or the projection? Without an ablation separating these design choices, the reader cannot judge whether the proposed 'dual orthogonal' mechanism is the source of the improvement.
minor comments (3)
  1. [Abstract] The term 'OOV' is introduced only implicitly via 'out-of-vocabulary'; define the abbreviation on first use or avoid it.
  2. [Abstract] The phrase 'weak at the start and end of denoising, when the process is most sensitive' is somewhat vague—'sensitive' to what? Clarify the rationale for the triangular schedule.
  3. [General] The manuscript as provided to the reviewer consists only of the abstract; the 'Full Text' section is empty. The report above is therefore based solely on the abstract. This gap should be resolved before any substantive review can occur.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identified: DOG is an empirical guidance method whose reported gains are based on external benchmarks, not on a fitted or self-referential prediction.

full rationale

On the basis of the available manuscript text (abstract), the central claim is an empirical improvement in handwritten text generation using a Dual Orthogonal Guidance (DOG) sampling strategy. The method constructs a guidance direction as the orthogonal projection of a negatively perturbed prompt onto the original positive prompt; this is a fixed geometric operation on conditioning embeddings, not a parameter fitted to reproduce the target result. The triangular schedule is a hyperparameter choice, and tuning such a schedule is standard practice rather than circular reasoning. There is no equation in the abstract that equates a predicted quantity to a fitted quantity, no load-bearing self-citation, and no invocation of an author-derived uniqueness theorem. The orthogonality assumption is mathematically under-justified in the abstract, but that is a support or correctness concern, not circularity. Evaluations are on external benchmarks (DiffusionPen and One-DM), so the results are not self-referential. Therefore no circular step is present.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on untested geometric properties of the diffusion conditioning space and on several hyperparameters. These are domain assumptions and tuning choices, not derived results, and their validity determines whether DOG works.

free parameters (2)
  • DOG guidance scale
    The strength of the orthogonal guidance is a tuning parameter, not derived from first principles. No value is given in the abstract.
  • Triangular schedule ramp positions
    The schedule is described as weak at start and end and strong in the middle, but the exact peak location and slope are not specified and are presumably tuned.
assumptions (3)
  • domain assumption The diffusion conditioning space supports linear operations such as subtraction and orthogonal projection, and these operations preserve semantic meaning.
    DOG's core operation is an orthogonal projection of a perturbed prompt onto the original prompt; for this to steer generation, embedding arithmetic must align with visual semantics.
  • domain assumption A negatively perturbed prompt encodes perturbations that cause artifacts rather than content changes.
    The guidance direction is derived from a negative prompt; this assumes the negative perturbation direction isolates unwanted features.
  • domain assumption The denoising process is most sensitive at the start and end, justifying the triangular schedule.
    The schedule is motivated by a sensitivity claim, not derived from a formal analysis. The paper states this as a premise for the schedule design.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual Orthogonal Guidance for Robust Diffusion-based Handwritten Text Generation." pith.science (2026). https://pith.science/paper/B6BPTDMI

@misc{pith2026250817017,
  author       = {Pith},
  title        = {Pith review of: Dual Orthogonal Guidance for Robust Diffusion-based Handwritten Text Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B6BPTDMI}},
  note         = {Machine review of arXiv:2508.17017}
}
read the original abstract

Diffusion-based Handwritten Text Generation (HTG) approaches achieve impressive results on frequent, in-vocabulary words observed at training time and on regular styles. However, they are prone to memorizing training samples and often struggle with style variability and generation clarity. In particular, standard diffusion models tend to produce artifacts or distortions that negatively affect the readability of the generated text, especially when the style is hard to produce. To tackle these issues, we propose a novel sampling guidance strategy, Dual Orthogonal Guidance (DOG), that leverages an orthogonal projection of a negatively perturbed prompt onto the original positive prompt. This approach helps steer the generation away from artifacts while maintaining the intended content, and encourages more diverse, yet plausible, outputs. Unlike standard Classifier-Free Guidance (CFG), which relies on unconditional predictions and produces noise at high guidance scales, DOG introduces a more stable, disentangled direction in the latent space. To control the strength of the guidance across the denoising process, we apply a triangular schedule: weak at the start and end of denoising, when the process is most sensitive, and strongest in the middle steps. Experimental results on the state-of-the-art DiffusionPen and One-DM demonstrate that DOG improves both content clarity and style variability, even for out-of-vocabulary words and challenging writing styles.

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. Diffusion-Based Ukrainian Handwritten Text Generation with Cross-Domain Style Transfer

    cs.CV 2026-05 unverdicted novelty 3.0 of 10

    Retraining a latent diffusion model on a newly constructed Ukrainian handwritten dataset enables cross-lingual and few-shot style transfer from English to Cyrillic, yielding legible and style-consistent word images.

Reference graph

Works this paper leans on

50 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Adversarial Generation of Handwritten Text Images Conditioned on Sequences

    Eloi Alonso, Bastien Moysset, and Ronaldo Messina. Adversarial Generation of Handwritten Text Images Conditioned on Sequences . In 2019 International Conference on Document Analysis and Recognition (ICDAR), pages 481--486. IEEE, 2019

    2019

  3. [3]

    Towards Principled Methods for Training Generative Adversarial Networks

    Martin Arjovsky and Léon Bottou. Towards Principled Methods for Training Generative Adversarial Networks . arXiv preprint arXiv:1701.04862, 2017

    arXiv 2017

  4. [4]

    Re-imagine the Negative Prompt Algorithm: Transform 2d Diffusion into 3D, alleviate Janus problem and Beyond

    Mohammadreza Armandpour, Ali Sadeghian, Huangjie Zheng, Amir Sadeghian, and Mingyuan Zhou. Re-imagine the Negative Prompt Algorithm: Transform 2d Diffusion into 3D, alleviate Janus problem and Beyond . arXiv preprint arXiv:2304.04968, 2023

    arXiv 2023

  5. [5]

    Handwriting Transformers

    Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Fahad Shahbaz Khan, and Mubarak Shah. Handwriting Transformers . In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 1086--1094, 2021

    2021

  6. [6]

    Semi-Supervised Adaptation of Diffusion Models for Handwritten Text Generation

    Kai Brandenbusch. Semi-Supervised Adaptation of Diffusion Models for Handwritten Text Generation . arXiv preprint arXiv:2412.15853, 2024

    work page Pith review arXiv 2024

  7. [7]

    Boosting Modern and Historical Handwritten Text Recognition with Deformable Convolutions

    Silvia Cascianelli, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. Boosting Modern and Historical Handwritten Text Recognition with Deformable Convolutions . International Journal on Document Analysis and Recognition (IJDAR), 25 0 (3): 0 207--217, 2022

    2022

  8. [8]

    Diffusion models in vision: A survey

    Florinel-Alin Croitoru, Vlad Hondru, Radu Tudor Ionescu, and Mubarak Shah. Diffusion models in vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 0 (9): 0 10850--10869, 2023

    2023

Show all 50 references
  1. [9]

    One-DM: One-Shot Diffusion Mimicker for Handwritten Text Generation

    Gang Dai, Yifan Zhang, Quhui Ke, Qiangya Guo, and Shuangping Huang. One-DM: One-Shot Diffusion Mimicker for Handwritten Text Generation . In European Conference on Computer Vision, pages 410--427. Springer, 2025

    2025

  2. [10]

    Davis, Chris Tensmeyer, Brian L

    Brian L. Davis, Chris Tensmeyer, Brian L. Price, Curtis Wigington, B. Morse, and R. Jain. Text and Style Conditioned GAN for the Generation of Offline-Handwriting Lines . In Proceedings of the 31^ st British Machine Vision Conference (BMVC) , 2020

    2020

  3. [11]

    A survey of handwriting synthesis from 2019 to 2024: A comprehensive review

    Moises Diaz, Andrea Mendoza-Garc \' a, Miguel A Ferrer, and Robert Sabourin. A survey of handwriting synthesis from 2019 to 2024: A comprehensive review . Pattern Recognition, page 111357, 2025

    2019

  4. [12]

    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 . In International Conference on Document Analysis and Recognition, pages 20--37. Springer, 2023

    2023

  5. [13]

    The Fr \'e chet distance between multivariate normal distributions

    DC Dowson and BV666017 Landau. The Fr \'e chet distance between multivariate normal distributions . Journal of Multivariate Analysis, 12 0 (3): 0 450--455, 1982

    1982

  6. [14]

    Do GAN s always have N ash equilibria? In International Conference on Machine Learning, pages 3029--3039

    Farzan Farnia and Asuman Ozdaglar. Do GAN s always have N ash equilibria? In International Conference on Machine Learning, pages 3029--3039. PMLR, 2020

    2020

  7. [15]

    HiGAN+: Handwriting Imitation GAN with Disentangled Representations

    Ji Gan, Weiqiang Wang, Jiaxu Leng, and Xinbo Gao. HiGAN+: Handwriting Imitation GAN with Disentangled Representations . ACM Transactions on Graphics (TOG), 42 0 (1): 0 1--17, 2022

    2022

  8. [16]

    Generating sequences with recurrent neural networks

    Alex Graves. Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850, 2013

    2013 arXiv

  9. [17]

    Word-Diffusion: Diffusion-Based Handwritten Text Word Image Generation

    Aniket Gurav, Narayanan C Krishnan, and Sukalpa Chanda. Word-Diffusion: Diffusion-Based Handwritten Text Word Image Generation . In International Conference on Pattern Recognition, pages 53--72. Springer, 2025

    2025

  10. [18]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-Free Diffusion Guidance . In NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2021

    2021

  11. [19]

    Denoising Diffusion Probabilistic Models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising Diffusion Probabilistic Models . Advances in Neural Information Processing Systems, 33: 0 6840--6851, 2020

    2020

  12. [20]

    ST-KeyS: Self-supervised Transformer for Keyword Spotting in Historical Handwritten Documents

    Sana Khamekhem Jemni, Sourour Ammar, Mohamed Ali Souibgui, Yousri Kessentini, and Abbas Cheddad. ST-KeyS: Self-supervised Transformer for Keyword Spotting in Historical Handwritten Documents . Pattern Recognition, page 112036, 2025

    2025

  13. [21]

    GANwriting: Content-Conditioned Generation of Styled Handwritten Word Images

    Lei Kang, Pau Riba, Yaxing Wang, Mar c al Rusinol, Alicia Forn \'e s, and Mauricio Villegas. GANwriting: Content-Conditioned Generation of Styled Handwritten Word Images . In European Conference on Computer Vision, pages 273--289. Springer, 2020

    2020

  14. [22]

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

    Lei Kang, Pau Riba, Marcal Rusinol, Alicia Forn \'e s, and Mauricio Villegas. Content and style aware generation of text-line images for handwriting recognition . IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    2021

  15. [23]

    Pay attention to what you read: non-recurrent handwritten text-line recognition

    Lei Kang, Pau Riba, Mar c al Rusi \ n ol, Alicia Forn \'e s, and Mauricio Villegas. Pay attention to what you read: non-recurrent handwritten text-line recognition. PR, 129: 0 108766, 2022

    2022

  16. [24]

    GNHK: A Dataset for English Handwriting in the Wild

    Alex WC Lee, Jonathan Chung, and Marco Lee. GNHK: A Dataset for English Handwriting in the Wild . In Document Analysis and Recognition--ICDAR 2021: 16th International Conference, Lausanne, Switzerland, September 5--10, 2021, Proceedings, Part IV 16, pages 399--412. Springer, 2021

    2021

  17. [25]

    Style-preserving english handwriting synthesis

    Zhouchen Lin and Liang Wan. Style-preserving english handwriting synthesis. Pattern Recognition, 40 0 (7): 0 2097--2109, 2007

  18. [26]

    Compositional Visual Generation with Composable Diffusion Models

    Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional Visual Generation with Composable Diffusion Models . In European Conference on Computer Vision, pages 423--439. Springer, 2022

    2022

  19. [27]

    Understanding diffusion models: A unified perspective

    Calvin Luo. Understanding diffusion models: A unified perspective. arXiv preprint arXiv:2208.11970, 2022

    2022 arXiv

  20. [28]

    The IAM-database: an English sentence database for offline handwriting recognition

    U-V Marti and Horst Bunke. The IAM-database: an English sentence database for offline handwriting recognition . International Journal on Document Analysis and Recognition, 5: 0 39--46, 2002

    2002

  21. [29]

    Smartpatch: Improving Handwritten Word Imitation with Patch Discriminators

    Alexander Mattick, Martin Mayr, Mathias Seuret, Andreas Maier, and Vincent Christlein. Smartpatch: Improving Handwritten Word Imitation with Patch Discriminators . In International Conference on Document Analysis and Recognition, pages 268--283. Springer, 2021

    2021

  22. [30]

    Zero-Shot Paragraph-level Handwriting Imitation with Latent Diffusion Models

    Martin Mayr, Marcel Dreier, Florian Kordon, Mathias Seuret, Jochen Z \"o llner, Fei Wu, Andreas Maier, and Vincent Christlein. Zero-Shot Paragraph-level Handwriting Imitation with Latent Diffusion Models . International Journal of Computer Vision, pages 1--22, 2025

    2025

  23. [31]

    Which training methods for GAN s do actually converge? In International Conference on Machine Learning, pages 3481--3490

    Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for GAN s do actually converge? In International Conference on Machine Learning, pages 3481--3490. PMLR, 2018

    2018

  24. [32]

    A Survey of Historical Document Image Datasets

    Konstantina Nikolaidou, Mathias Seuret, Hamam Mokayed, and Marcus Liwicki. A Survey of Historical Document Image Datasets . International Journal on Document Analysis and Recognition (IJDAR), 25 0 (4): 0 305--338, 2022

    2022

  25. [33]

    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 . In International Conference on Document A...

    2023

  26. [34]

    DiffusionPen: Towards Controlling the Style of Handwritten Text Generation

    Konstantina Nikolaidou, George Retsinas, Giorgos Sfikas, and Marcus Liwicki. DiffusionPen: Towards Controlling the Style of Handwritten Text Generation . In European Conference on Computer Vision, pages 417--434. Springer, 2024 a

    2024

  27. [35]

    Rethinking HTG Evaluation: Bridging Generation and Recognition

    Konstantina Nikolaidou, George Retsinas, Giorgos Sfikas, and Marcus Liwicki. Rethinking HTG Evaluation: Bridging Generation and Recognition . In European Conference on Computer Vision, pages 179--195. Springer, 2024 b

    2024

  28. [36]

    Handwritten Text Generation from Visual Archetypes

    Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Handwritten Text Generation from Visual Archetypes . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22458--22467, 2023 a

    2023

  29. [37]

    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 . In BMVC, 2023 b

    2023

  30. [38]

    Keyword Spotting in Handwritten Documents Using Projections of Oriented Gradients

    George Retsinas, Georgios Louloudis, Nikolaos Stamatopoulos, and Basilis Gatos. Keyword Spotting in Handwritten Documents Using Projections of Oriented Gradients . In 2016 12th IAPR Workshop on Document Analysis Systems (DAS), pages 411--416. IEEE, 2016

    2016

  31. [39]

    Efficient Learning-Free Keyword Spotting

    George Retsinas, Georgios Louloudis, Nikolaos Stamatopoulos, and Basilis Gatos. Efficient Learning-Free Keyword Spotting . IEEE Transactions on Pattern Analysis and Machine Intelligence, 41 0 (7): 0 1587--1600, 2018

    2018

  32. [40]

    From Seq2Seq Recognition to Handwritten Word Embeddings

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

    2021

  33. [41]

    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. In International Workshop on Document Analysis Systems, pages 247--259. Springer, 2022

    2022

  34. [42]

    Enhancing CRNN HTR Architectures with Transformer Blocks

    George Retsinas, Konstantina Nikolaidou, and Giorgos Sfikas. Enhancing CRNN HTR Architectures with Transformer Blocks . In International Conference on Document Analysis and Recognition, pages 425--440. Springer, 2024

    2024

  35. [43]

    Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models

    Seyedmorteza Sadat, Otmar Hilliges, and Romann M Weber. Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models . In The Thirteenth International Conference on Learning Representations, 2025

    2025

  36. [44]

    Scrabble GAN : Semi-Supervised Varying Length Handwritten Text Generation

    Sharon Fogel and Hadar Averbuch-Elor and Sarel Cohen and Shai Mazor and Roee Litman . Scrabble GAN : Semi-Supervised Varying Length Handwritten Text Generation . 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 4323--4332 , 2020

    2020

  37. [45]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, pages 2256--2265. PMLR, 2015

    2015

  38. [46]

    Denoising Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising Diffusion Implicit Models . In ICLR, 2021

    2021

  39. [47]

    Synthetic handwritten captchas

    Achint Oommen Thomas, Amalia Rusu, and Venu Govindaraju. Synthetic handwritten captchas. Pattern Recognition, 42 0 (12): 0 3365--3373, 2009

    2009

  40. [48]

    VATr++: Choose Your Words Wisely for Handwritten Text Generation

    Bram Vanherle, Vittorio Pippi, Silvia Cascianelli, Nick Michiels, Frank Van Reeth, and Rita Cucchiara. VATr++: Choose Your Words Wisely for Handwritten Text Generation . IEEE Transactions on Pattern Analysis and Machine Intelligence, 47 0 (2): 0 934--948, 2025

    2025

  41. [49]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 13 0 (4): 0 600--612, 2004

    2004

  42. [50]

    Conditional Text Image Generation with Diffusion Models

    Yuanzhi Zhu, Zhaohai Li, Tianwei Wang, Mengchao He, and Cong Yao. Conditional Text Image Generation with Diffusion Models . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14235--14245, 2023

    2023

Pith tools

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