Pith. sign in

REVIEW 4 major objections 3 minor 1 cited by

On the Generalization of Handwritten Text Recognition Models

T0 review · 4 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Handwritten text recognition fails out of distribution mainly because the target words and characters differ from the source, not because the handwriting looks unfamiliar.

desk verdict A valuable OOD benchmark for HTR with a robust raw gap, but the factor-ordering and label-free estimator claims outrun the evidence. read the letter →

arxiv 2411.17332 v2 pith:T5XIFLBM submitted 2024-11-26 cs.LG

Carlos Garrido-Munoz , Jorge Calvo-Zaragoza This is my paper · ORCID
classification cs.LG
keywords handwrittentextrecognitiondomaingeneralizationout-of-distributionerrortextualdivergencevisualfactoranalysissyntheticdatacharacterrate
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 (HTR) models are usually trained and tested on the same distribution, but real documents arrive from unknown scripts, eras, and languages. This paper runs the first large-scale domain-generalization study for HTR, with 336 out-of-distribution evaluations across eight architectures, seven datasets, and five languages, and asks what actually breaks these models. Its central claim is that the dominant cause of out-of-distribution failure is textual divergence between source and target — the characters and words the model must emit — with visual divergence playing a secondary role. The paper further claims that out-of-distribution error can be estimated in advance from proxies that do not require target labels, with an average error of about 10.9 CER points and roughly 70% of predictions falling within 10 points. If this is right, practitioners can forecast which domains will break their recognizer before deploying it.

What carries the argument

The argument rides on two novel divergence metrics and a factor analysis. Visual divergence Δ(X_S, X_T) is the average reconstruction error (MSE) of a convolutional autoencoder trained on source images when it is fed target images, a reconstruction-based anomaly-detection score. Textual divergence Δ(Y_S, Y_T) is the averaged KL divergence, over n = 1 to 5, between the character n-gram distributions of source and target text; the paper distinguishes Δ_L, the divergence between the source text and synthetic text of a known language, from Δ_GT, the divergence between the source text and the ground-truth target transcriptions. These metrics feed a factor analysis (eigenvalues ≥ 1, oblimax rotation) producing four latent factors, and the resulting factor structure, together with model capacity and calibration metrics, forms the input to the out-of-distribution error estimator.

What would settle it

Re-run the Section 5.2 estimator on the same 336 cases with Δ_L, the synthetic language-matched divergence, as the only textual input and compute the mean absolute error; if it exceeds roughly 11 CER points, or fewer than 70% of residuals fall below 10 points, the headline predictive claim fails. As a simpler check, inspect the estimator's feature list: if Δ_GT appears among the inputs, the 70%-below-10-points statistic is not a label-free prediction.

Watch

Extended reading notes

Core claim

Trained from scratch on a single source domain and evaluated on every other domain, eight state-of-the-art HTR models collapse out of distribution: average CER rises from roughly 5–10% in distribution to 37–54% out of distribution. A factor analysis over model parameters, in-distribution and out-of-distribution CER and calibration error, visual divergence, and textual divergence extracts four latent factors, and the factor loaded on textual divergence — averaged KL divergence over 1-to-5-gram character distributions between source and target text — correlates most strongly with out-of-distribution error (0.62). The visual-divergence factor, measured as autoencoder reconstruction error, shows only a weak secondary influence, with a negative loading on OOD error (−0.3). The paper concludes that textual divergence is the most significant factor for generalization, followed by visual divergence, and demonstrates that OOD error can be estimated from proxy metrics without target labels, reaching a mean absolute error of 10.9 CER points.

Load-bearing premise

The claim that out-of-distribution error can be forecast to roughly 11 CER points depends on which textual-divergence measure feeds the estimator: if it is the ground-truth divergence Δ_GT, the forecast smuggles in the very target transcriptions that domain generalization forbids access to, and the paper never shows that the label-free divergence Δ_L alone achieves the claimed accuracy.

Editorial extensions

If this is right

  • No current HTR architecture generalizes acceptably: the best out-of-distribution CER is 37.4%, and the average in-distribution-to-out-of-distribution gap is 37.6 CER points, so benchmark gains under the i.i.d. assumption do not transfer to unseen domains.
  • CTC-aligned models are the safer choice for generalization (average OOD CER 40.4 vs. 48.7 for autoregressive decoders), and the internal language model of autoregressive models biases them more strongly toward the source language.
  • Synthetic training data improves out-of-distribution performance even when rendered in a different language — about 6 CER points on average, and over 11 when the synthetic language matches the target — making synthetic data a cheap lever for generalization.
  • Model selection strategy and model capacity have negligible impact on out-of-distribution performance, so tuning those will not close the generalization gap in HTR.
  • Because textual divergence dominates, out-of-distribution error can be estimated in advance from proxies, with roughly 70% of estimates within 10 CER points, allowing risky target domains to be flagged before deployment.

Reading between the lines

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

  • The negative loading of the visual factor on OOD error (−0.3) sits awkwardly with the abstract's ordering of the two factors; if it holds, a recognizer fails most where the text differs even when the handwriting looks similar, meaning visual novelty alone is a poor basis for predicting failure.
  • The estimator's accuracy across 336 cases in five languages suggests the same proxy recipe — in-distribution error, calibration, reconstruction error, and n-gram KL divergence — could transfer to other sequence-recognition tasks such as automatic speech recognition or OCR, where target transcripts are also unavailable at deployment.
  • A direct testable extension is to feed the estimator only Δ_L, the synthetic language-matched divergence, and check whether the 70%-below-10-points accuracy survives; if it does, the predictor is genuinely label-free, and if not, the headline predictive claim needs an explicit label-free baseline.
  • The finding that autoregressive models are more biased by source language implies that hybrid CTC models, or decoders whose language model can be swapped at test time, are the more promising route for cross-lingual HTR.
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 / 3 minor

Summary. The paper presents a large-scale empirical study of out-of-distribution (OOD) generalization for handwritten text recognition (HTR). Eight HTR architectures are trained on seven real datasets across five languages, yielding 336 OOD evaluations, plus additional experiments with synthetic text rendered in five languages. The authors report that OOD error is much higher than in-distribution error, that CTC-based alignment generalizes best, that synthetic data give a modest OOD improvement, and that model selection and capacity have little effect. They then perform a factor analysis over features that include visual and textual divergence proxies, concluding that textual divergence is the most significant factor and visual divergence is second. Finally, they claim that OOD error can be estimated from label-free proxies with a mean absolute error of 10.9 CER points and that about 70% of predictions fall below 10 points.

Significance. If the claims are supported, the paper would be a valuable first systematic study of HTR under the domain-generalization setting, with practical guidance for practitioners and a useful public benchmark. The raw empirical finding—that OOD CER is dramatically higher than ID CER across all architectures and datasets—is robust and consistently presented in Tables 3–7, and the release of code is a concrete asset. The synthetic-data comparison and the analysis of model-selection strategies are also informative. However, the two headline claims—the factor ordering and the label-free OOD-error estimator—are not yet established by the evidence as presented. The factor-ordering conclusion conflicts with the displayed loadings, and the estimator section leaves unspecified whether the KL-divergence input uses ground-truth target transcriptions or a synthetic proxy, which is decisive for the label-free claim. These issues affect the central contributions and require a revision.

major comments (4)
  1. [Section 5.2] The estimator described in Section 5.2 is said to use 'metrics that do not require target labels... and KL divergence', but the manuscript does not specify which KL divergence is used. Section 5.1 defines two textual divergences: ΔGT(YS,YT), computed from ground-truth target transcriptions, and ΔL(YS,YT), computed against a synthetic corpus. If ΔGT enters the estimator, the label-free claim is false and the reported 70%-below-10-points result partly measures leakage of the target labels that are also used to compute the OOD CER. If ΔL is used instead, the paper provides no ablation or standalone result showing that a ΔL-only estimator achieves the claimed accuracy; moreover, selecting the synthetic corpus requires knowing the target language, which the DG setup in Section 3.3 explicitly excludes. Please specify which divergence is used and provide a separate evaluation for the truly label-free version.
  2. [Section 5.1, Fig. 6] The abstract's ordering claim—'textual divergence... followed by visual divergence'—is not supported by the displayed factor loadings. Factor 2, labeled visual divergence, loads 0.80 on ΔS and 0.92 on ΔT but has a negative loading of -0.30 on OOD error, so the visual factor is not a positive driver of OOD error in this analysis. Furthermore, because OOD error is itself an input to the factor analysis, the loadings describe shared variance rather than a causal or predictive ordering; no regression or variance decomposition is reported that would quantify the relative contribution of textual versus visual divergence to OOD error. The conclusion in Section 6 appropriately hedges ('We emphasize that more research studying this last factor has to be done'), but the abstract does not, and the two should be made consistent.
  3. [Section 5.2, Fig. 7] The OOD-error estimator is described only as 'we calculated the expected error'; the manuscript does not state what model is fit, how the coefficients are obtained, or whether the evaluation is held-out or cross-validated. If the estimator is fit and evaluated on the same 336 cases, the reported MAE of 10.9 and the cumulative residual distribution are training-set diagnostics, and the phrase 'can be reliably estimated' overstates predictive validity. Please describe the fitting procedure and report a held-out or cross-validated evaluation.
  4. [Abstract, Section 5.2, Fig. 7] The claimed threshold is numerically inconsistent with the cumulative plot. In Fig. 7 (right), the cumulative percentages at x-axis positions 0, 5, 10, 15 are approximately 29.1, 50.8, 68.3, 85.2, which implies that about 50.8% of residuals are below 10 CER points and about 68.3% are below 15 points. The abstract's 'discrepancies falling below 10 points in 70% of cases' and the Fig. 7 caption's 'below 10 points of CER' do not match this plot. Please correct the text or clarify the binning; if the intended statement is 'below 15 points', the abstract and caption must be revised accordingly.
minor comments (3)
  1. [Table 3 and Section 4.1] The averages in Table 3 are computed after filtering out non-converged runs marked with a star, but no sensitivity analysis is provided. Since non-converged runs are concentrated in particular architectures, please report whether the qualitative conclusions (e.g., the ranking of alignment types) are robust to including or excluding these runs.
  2. [Fig. 7 caption and Section 5.2 text] The figure caption reports 'MSE of 10.9 on average' while the main text reports the same numeric value as Mean Absolute Error (MAE); both cannot be correct for the same quantity. Please correct the metric name and value.
  3. [Appendix 7] The paragraph after Table 5 ends with an incomplete sentence beginning 'Additionally,'; this appears to be a leftover fragment and should be removed or completed.

Circularity Check

2 steps flagged · score 6.0 of 10

The Section 5.2 OOD-error estimator is not a validated prediction: its 'KL divergence' input is ambiguous between ΔGT (leaking target transcripts) and ΔL (unvalidated without ablation), and the reported 70% within-10-point accuracy is computed without any described held-out split, so the central predictive claim reduces to an in-sample fit or label leakage.

  1. other [Section 5.1 and Section 5.2]
    "We refer to ΔGT (YS, YT ) as the actual divergence (see Appendix 9) between the source domain YS and the ground truth of the target domain YT . ... To analyze this, we calculated the expected error using metrics that do not require target labels: ID error, ID ECE, number of parameters, reconstruction errors (ID and OOD), and KL divergence."

    The estimator's input list is advertised as label-free, but 'KL divergence' is not specified. Section 5.1 defines two KL-based quantities: ΔL(YS,YT), computed against a synthetic corpus in a known language, and ΔGT(YS,YT), computed against the ground-truth transcriptions of the target domain. Those ground-truth transcriptions are exactly the labels used to compute the actual CER that the estimator claims to predict. If the KL divergence fed to the estimator is ΔGT, then the target labels enter both the predictor and the predicted quantity, so the reported accuracy is partly a consequence of label leakage rather than of a label-free proxy.

  2. fitted input called prediction [Section 5.2 and Fig. 7]
    "The results are displayed in Fig. 7 (left), where the actual CER (X-axis) is plotted against the expected CER (Y-axis). ... Right: Cumulative distribution of grouped residuals. Approximately 70% of predictions yield an error below 10 points of CER."

    The paper does not describe any train/validation split, cross-validation, or external benchmark for the expected-error model. The 336 OOD cases used to construct the estimator appear to be the same 336 cases on which the residuals are measured. The 70% figure is therefore an in-sample accuracy statement, not a predictive result. Calling the fitted regression's residuals 'predictions' renames an in-sample fit as validated forecasting, and no evidence is provided that the estimator would achieve comparable residuals on unseen OOD cases.

full rationale

The paper's main empirical construction—336 OOD evaluations, factor analysis, and synthetic-data comparisons—is largely self-contained and does not rely on self-citation or on redefining known results. The factor analysis is descriptive, and the visual/textual divergence metrics are computed from data rather than from the OOD error itself, so the factor loadings are not circular. The circularity problem is concentrated in Section 5.2. The estimator's input list is claimed to be label-free, yet 'KL divergence' is ambiguous between ΔL (synthetic corpus, requiring target-language knowledge) and ΔGT (ground-truth target transcriptions, which are the same labels used to score CER). If ΔGT is used, the 70% below-10-point claim partly reflects label leakage; if only ΔL is used, no ablation supports the claim. In either case, no held-out split is reported for the expected-error model, so the residual distribution is an in-sample fit presented as prediction. The conclusions also hedge that more research on the visual factor is needed, and Fig. 6 shows Factor 2 (visual divergence) loading negatively on OOD error (-0.3), which weakens the abstract's ordering but is more a correctness concern than a circularity concern. Overall, the central predictive claim is not independently validated; score 6 reflects partial circularity in the estimator's construction and evaluation.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central empirical claims rely on proxy metrics that are treated as measurements, plus an estimator whose coefficients are never disclosed. The most important unaccounted degree of freedom is the OOD-error estimator: it is fit to the same data used for evaluation, and no held-out validation is described. The factor analysis adds a data-driven choice k equals 4. No new physical entities are introduced.

free parameters (4)
  • OOD-error estimator coefficients = not reported
    The estimator plotted in Fig. 7 is never written down; it is fit to the observed OOD CER values and then evaluated on the same data, so its coefficients are hidden free parameters of the headline '70%' claim.
  • Number of retained factors in factor analysis = 4
    Chosen by the eigenvalue at least 1 rule on the standardized data matrix. This data-driven modeling choice shapes which latent factor is labeled 'textual divergence' and which is labeled 'visual divergence'.
  • Synthetic data generation budget = 100,000 lines, 4,000 fonts, equal lines per language
    The synthetic-data conclusions depend on this hand-picked budget. Other budgets could change the 'synthetic data slightly helps' result and the relative ranking of models.
  • n-gram averaging range for textual divergence = n = 1 to 5
    Averaging KL divergences over n-grams 1 to 5 is a chosen hyperparameter of the main explanatory metric. The paper does not test sensitivity to this range.
assumptions (5)
  • standard math Factor analysis with oblimax rotation gives interpretable latent factors that can be used to infer which input variables explain OOD error.
    Used in Section 5.1 to conclude that Factor 1 is 'textual divergence' and that it is the most significant contributor to OOD error. Factor analysis only identifies latent structure in the included variables and does not by itself establish causal or predictive ordering among them.
  • domain assumption Average reconstruction error of a per-source convolutional autoencoder is a valid proxy for visual divergence between handwriting domains.
    Section 5 and Appendix 9. This proxy is never validated against human judgments or independent distribution-shift measures, and it yields a negative correlation with OOD error in the factor analysis, which is not reconciled with the paper's prose.
  • domain assumption Average KL divergence over n-grams between source and target text is a valid proxy for textual divergence.
    Section 5. If implemented as delta_GT, it requires ground-truth transcriptions of the target domain, which are unavailable in the DG setting. If implemented as delta_L, it measures divergence to synthetic text in a known language, which is a weaker object than actual target text divergence.
  • ad hoc to paper Non-converged runs marked with a star can be excluded when computing average OOD performance.
    Table 3 footnotes state that starred results are outliers because the model did not converge in the ID setting, and averages are computed filtering out outliers. This exclusion changes the rank ordering of models, for example HTR-VT on G.W., and is applied without a pre-specified convergence criterion.
  • domain assumption The seven datasets can be treated as independent domains with no accidental writer overlap or hidden leakage between train and test across datasets.
    The entire 336-case OOD matrix treats each dataset as a distinct domain. No intersection of writers or documents across datasets is verified in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the Generalization of Handwritten Text Recognition Models." pith.science (2026). https://pith.science/paper/T5XIFLBM

@misc{pith2026241117332,
  author       = {Pith},
  title        = {Pith review of: On the Generalization of Handwritten Text Recognition Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T5XIFLBM}},
  note         = {Machine review of arXiv:2411.17332}
}
read the original abstract

Recent advances in Handwritten Text Recognition (HTR) have led to significant reductions in transcription errors on standard benchmarks under the i.i.d. assumption, thus focusing on minimizing in-distribution (ID) errors. However, this assumption does not hold in real-world applications, which has motivated HTR research to explore Transfer Learning and Domain Adaptation techniques. In this work, we investigate the unaddressed limitations of HTR models in generalizing to out-of-distribution (OOD) data. We adopt the challenging setting of Domain Generalization, where models are expected to generalize to OOD data without any prior access. To this end, we analyze 336 OOD cases from eight state-of-the-art HTR models across seven widely used datasets, spanning five languages. Additionally, we study how HTR models leverage synthetic data to generalize. We reveal that the most significant factor for generalization lies in the textual divergence between domains, followed by visual divergence. We demonstrate that the error of HTR models in OOD scenarios can be reliably estimated, with discrepancies falling below 10 points in 70\% of cases. We identify the underlying limitations of HTR models, laying the foundation for future research to address this challenge.

Figures

Figures reproduced from arXiv: 2411.17332 by the authors.

Figure 1
Figure 1. Average performance of HTR models for in-distribution [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the different approaches for the concept of “generalization” in HTR literature. While the field commonly considers [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Average CER in the out-of-distribution (OOD) scenario [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Average CER (%) in out-of-distribution (OOD) perfor [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Left: Representation of the estimated vs. real CER val [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: Heatmap of textual divergence from real HTR domains. [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Heatmap of textual divergence between real and syn [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 8
Figure 8. Figure 8: Heatmap of visual divergence between source (rows) [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 11
Figure 11. Figure 11: Representation of visual divergence between domains. Examples are presented in two columns: on the left, three domain pairs [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: Scree plot: Eigenvalues of the standardized values used [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Handwritten Text Recognition: A Survey

    cs.CV 2025-02 conditional novelty 3.0 of 10

    A survey of handwritten text recognition that categorizes methods by reading-order complexity and compares reported performance on the IAM database.

Reference graph

Works this paper leans on

103 extracted references · 74 canonical work pages · cited by 1 Pith paper

  1. [1]

    Hamada, and Daniyar Nurseitov

    Abdelrahman Abdallah, Mohamed A. Hamada, and Daniyar Nurseitov. Attention-based fully gated cnn-bgru for russian handwritten text. Journal of Imaging, 2020. 2

    2020

  2. [2]

    International Conference on Frontiers in Handwriting Recognition (ICFHR 2010) - Competitions Overview

    Haikal El Abed, V olker M ¨argner, and Michael Blumen- stein. International Conference on Frontiers in Handwriting Recognition (ICFHR 2010) - Competitions Overview. 2010. 1, 2

    2010

  3. [3]

    Manmatha, and Pietro Per- ona

    Aviad Aberdam, Ron Litman, Shahar Tsiper, Oron Anschel, Ron Slossberg, Shai Mazor, R. Manmatha, and Pietro Per- ona. Sequence-to-sequence contrastive learning for text recognition. Computer Vision and Pattern Recognition ,

  4. [4]

    Jose Carlos Aradillas, Juan Jose Murillo-Fuentes, and Pablo M. Olmos. Boosting Offline Handwritten Text Recog- nition in Historical Documents With Few Labeled Lines. IEEE Access, 9:76674–76688, 2021. 3

    2021

  5. [5]

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

    2021

  6. [6]

    Jose Carlos Aradillas Jaramillo, Juan Jose Murillo-Fuentes, and Pablo M. Olmos. Boosting Handwriting Text Recog- nition in Small Databases with Transfer Learning. In 2018 16th International Conference on Frontiers in Handwriting Recognition (ICFHR) , pages 429–434, Niagara Falls, NY , USA, 2018. IEEE. 2, 3

    2018

  7. [7]

    Castellanos, and Jorge Calvo- Zaragoza

    Eric Ayllon, Francisco J. Castellanos, and Jorge Calvo- Zaragoza. Analysis of the calibration of handwriting text recognition models. In Document Analysis and Recognition - ICDAR 2024 , pages 139–155, Cham, 2024. Springer Na- ture Switzerland. 4

    2024

  8. [8]

    Agreement-on-the-line: Predicting the performance of neural networks under distribution shift

    Christina Baek, Yiding Jiang, Aditi Raghunathan, and Zico Kolter. Agreement-on-the-line: Predicting the performance of neural networks under distribution shift. Neural Informa- tion Processing Systems, 2022. 2

    2022

Show all 103 references
  1. [9]

    Neural machine translation by jointly learning to align and translate

    Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. In 3rd International Conference on Learning Rep- resentations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. 2, 3

    2015

  2. [10]

    A light transformer-based architecture for handwritten text recognition

    Killian Barrere, Yann Soullard, Aur ´elie Lemaitre, and Bertrand Co ¨uasnon. A light transformer-based architecture for handwritten text recognition. 2022. 1, 3, 4, 5, 2

    2022

  3. [11]

    Training transformer architectures on few annotated data: an application to historical handwritten text recognition

    Killian Barrere, Yann Soullard, Aur ´elie Lemaitre, and Bertrand Co ¨uasnon. Training transformer architectures on few annotated data: an application to historical handwritten text recognition. International Journal on Document Analy- sis and Recognition (IJDAR), 2024. 1, 3, 4, 5, 2

    2024

  4. [12]

    Sara Beery, Grant Van Horn, and P. Perona. Recognition in terra incognita. European Conference on Computer Vision,

  5. [13]

    A theory of learning from different domains

    Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. A theory of learning from different domains. Machine- mediated learning, 2010. 1

    2010

  6. [14]

    Paul Bergmann, Sindy L ¨owe, Michael Fauser, David Satt- legger, and C. Steger. Improving unsupervised defect seg- mentation by applying structural similarity to autoencoders. VISIGRAPP, 2018. 7

    2018

  7. [15]

    Paul Bergmann, Michael Fauser, David Sattlegger, and C. Steger. Mvtec ad — a comprehensive real-world dataset for unsupervised anomaly detection. Computer Vision and Pat- tern Recognition, 2019. 7

    2019

  8. [16]

    MetaHTR: Towards Writer-Adaptive Handwritten Text Recognition

    Ayan Kumar Bhunia, Shuvozit Ghose, Amandeep Ku- mar, Pinaki Nath Chowdhury, Aneeshan Sain, and Yi-Zhe Song. MetaHTR: Towards Writer-Adaptive Handwritten Text Recognition. In 2021 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 15825– 15834, Nashv...

    2021

  9. [17]

    Handwriting Transformers

    Ankan Kumar Bhunia, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Fahad Shahbaz Khan, and Mubarak Shah. Handwriting Transformers . In 2021 IEEE/CVF In- ternational Conference on Computer Vision (ICCV) , pages 1066–1074, Los Alamitos, CA, USA, 2021. IEEE Computer Society. 1, 6

    2021

  10. [18]

    Text is Text, No Matter What: Uni- fying Text Recognition using Knowledge Distillation

    Ayan Kumar Bhunia, Aneeshan Sain, Pinaki Nath Chowd- hury, and Yi-Zhe Song. Text is Text, No Matter What: Uni- fying Text Recognition using Knowledge Distillation. In 2021 IEEE/CVF International Conference on Computer Vi- sion (ICCV), pages 963–972, Montreal, QC, Canada, 2021. IEEE. 3

    2021

  11. [19]

    Building a volunteer com- munity: Results and findings from transcribe bentham

    Tim Causer and Valerie Wallace. Building a volunteer com- munity: Results and findings from transcribe bentham. Dig- ital Humanities Quarterly, 6(2), 2012. 4

    2012

  12. [20]

    End-to-end handwritten paragraph text recognition using a vertical attention network

    Denis Coquenet, Clement Chatelain, and Thierry Paquet. End-to-end handwritten paragraph text recognition using a vertical attention network. 2022. 1, 2, 3, 4, 5

    2022

  13. [21]

    Dan: a segmentation-free document attention network for handwritten document recognition

    Denis Coquenet, Cl ´ement Chatelain, and Thierry Paquet. Dan: a segmentation-free document attention network for handwritten document recognition. 2023. 3

    2023

  14. [22]

    Instructblip: towards general-purpose vision-language models with instruction tuning

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. Instructblip: towards general-purpose vision-language models with instruction tuning. In Proceed- ings of the 37th International Conference on Neura...

    2023

  15. [23]

    Dehak, P

    N. Dehak, P. Kenny, R ´eda Dehak, P. Dumouchel, and P. Ouellet. Front-end factor analysis for speaker verification. IEEE Transactions on Audio, Speech, and Language Pro- cessing, 2011. 2, 7 9

    2011

  16. [24]

    Rethinking text line recog- nition models

    Daniel Hernandez Diaz, Reeve Ingle, Siyang Qin, Alessan- dro Bissacco, and Yasuhisa Fujii. Rethinking text line recog- nition models. arXiv, 2021. 3

    2021

  17. [25]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2010 arXiv

  18. [26]

    Self-attention networks for non-recurrent hand- written text recognition

    Rafael d’Arce, Terence Norton, Sion Hannuna, and Nello Cristianini. Self-attention networks for non-recurrent hand- written text recognition. In Frontiers in Handwriting Recog- nition, pages 389–403. Springer International Publishing,

  19. [27]

    Varma, Khaled Kamal Saab, Jean- Benoit Delbrouck, Christopher Lee-Messer, Jared A

    Sabri Eyuboglu, M. Varma, Khaled Kamal Saab, Jean- Benoit Delbrouck, Christopher Lee-Messer, Jared A. Dunn- mon, James Y . Zou, and Christopher R’e. Domino: Discov- ering systematic errors with cross-modal embeddings. In- ternational Conference on Learning Representations , 2022. 1

    2022

  20. [28]

    Dtrocr: Decoder-only transformer for opti- cal character recognition

    Masato Fujitake. Dtrocr: Decoder-only transformer for opti- cal character recognition. arXiv.org, 2023. 3

    2023

  21. [29]

    On the limitations of general purpose domain generalisation methods, 2024

    Henry Gouk, Ondrej Bohdal, Da Li, and Timothy Hospedales. On the limitations of general purpose domain generalisation methods, 2024. 2, 6

    2024

  22. [30]

    Evandro Gouv ˆea and Marelie H. Davel. Kullback-leibler divergence-based ASR training data selection. In 12th An- nual Conference of the International Speech Communication Association, INTERSPEECH 2011, Florence, Italy, August 27-31, 2011, pages 2297–2300. ISCA, 2011. 7

    2011

  23. [31]

    Framewise phoneme classification with bidirectional lstm and other neural net- work architectures

    Alex Graves and J ¨urgen Schmidhuber. Framewise phoneme classification with bidirectional lstm and other neural net- work architectures. Neural Networks, 18(5):602–610, 2005. IJCNN 2005. 2

    2005

  24. [32]

    Graves, Santiago Fern ´andez, Faustino J

    A. Graves, Santiago Fern ´andez, Faustino J. Gomez, and J. Schmidhuber. Connectionist temporal classification: la- belling unsegmented sequence data with recurrent neural networks. ICML, 2006. 2, 3

    2006

  25. [33]

    In search of lost do- main generalization

    Ishaan Gulrajani and David Lopez-Paz. In search of lost do- main generalization. International Conference on Learning Representations, 2021. 2, 3, 6, 7

    2021

  26. [34]

    Long short-term memory

    Sepp Hochreiter and J ¨urgen Schmidhuber. Long short-term memory. Neural Comput., 9(8):1735–1780, 1997. 2

    1997

  27. [35]

    Reeve Ingle, Yasuhisa Fujii, Thomas Deselaers, Jonathan Baccash, and Ashok C

    R. Reeve Ingle, Yasuhisa Fujii, Thomas Deselaers, Jonathan Baccash, and Ashok C. Popat. A Scalable Handwritten Text Recognition System . In 2019 International Conference on Document Analysis and Recognition (ICDAR), pages 17–24. IEEE Computer Society, 2019. 1, 2

    2019

  28. [36]

    Synthetic data and artificial neural net- works for natural scene text recognition

    Max Jaderberg, Karen Simonyan, Andrea Vedaldi, and An- drew Zisserman. Synthetic data and artificial neural net- works for natural scene text recognition. arXiv: Computer Vision and Pattern Recognition, 2014. 1

    2014

  29. [37]

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

    Lei Kang, Pau Riba, Marc ¸al Rusi ˜nol, Alicia Forn ´es, and Mauricio Villegas. Pay attention to what you read: Non- recurrent handwritten text-line recognition. arXiv (Cornell University), 2020. 3, 4

    2020

  30. [38]

    Unsupervised Adaptation for Synthetic- to-Real Handwritten Word Recognition

    Lei Kang, Marcal Rusinol, Alicia Fornes, Pau Riba, and Mauricio Villegas. Unsupervised Adaptation for Synthetic- to-Real Handwritten Word Recognition. In 2020 IEEE Win- ter Conference on Applications of Computer Vision (WACV), pages 3491–3500, Snowmass Village, CO, USA, 2020. ...

    2020

  31. [39]

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

    Lei Kang, Pau Riba, Marc ¸al Rusi ˜nol, Alicia Forn ´es, and Mauricio Villegas. Pay attention to what you read: Non- recurrent handwritten text-line recognition. Pattern Recog- nition, 129:108766, 2022. 1, 4, 5, 2, 3

    2022

  32. [40]

    Attentionhtr: Handwritten text recognition based on attention encoder-decoder networks

    Dmitrijs Kass and Ekta Vats. Attentionhtr: Handwritten text recognition based on attention encoder-decoder networks

  33. [41]

    Handwrit- ten mail classification experiments with the rimes database

    Christopher Kermorvant and J ´erˆome Louradour. Handwrit- ten mail classification experiments with the rimes database. In International Conference on Frontiers in Handwriting Recognition, ICFHR 2010, Kolkata, India, 16-18 November 2010, pages 241–246. IEEE Computer Society, 2010. 4

    2010

  34. [42]

    Joint ctc- attention based end-to-end speech recognition using multi- task learning

    Suyoun Kim, Takaaki Hori, and Shinji Watanabe. Joint ctc- attention based end-to-end speech recognition using multi- task learning. In 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 4835–4839, 2017. 4

    2017

  35. [43]

    Towards Writing Style Adaptation in Handwriting Recognition, 2023

    Jan Koh ´ut, Michal Hradiˇs, and Martin Kiˇsˇs. Towards Writing Style Adaptation in Handwriting Recognition, 2023. Version Number: 1. 2, 3

    2023

  36. [44]

    Praveen Krishnan and C. V . Jawahar. Generating synthetic data for text recognition. arXiv: Computer Vision and Pat- tern Recognition, 2016. 1

    2016

  37. [45]

    Lexi- con and attention based handwritten text recognition system

    Lalita Kumari, Sukhdeep Singh, Vaibhav Varish Singh Rathore, Anuj Sharma, Lalita Kumari, Sukhdeep Singh, Vaibhav Varish Singh Rathore, and Anuj Sharma. Lexi- con and attention based handwritten text recognition system

  38. [46]

    Trocr: Transformer-based optical character recognition with pre-trained models

    Minghao Li, Tengchao Lv, Jingye Chen, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, and Furu Wei. Trocr: Transformer-based optical character recognition with pre-trained models. Proceedings of the ... AAAI Conference on Artificial Intelligence, 2023. 1, 3, 5

    2023

  39. [47]

    HTR-VT: Handwritten text recognition with vision trans- former

    Yuting Li, Dexiong Chen, Tinglong Tang, and Xi Shen. HTR-VT: Handwritten text recognition with vision trans- former. 158:110967, 2024. 1, 3, 4, 5, 2

    2024

  40. [48]

    Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge, 2024. 5, 1

    2024

  41. [49]

    Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks

    Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. Neural Information Processing Systems, 2019. 3

    2019

  42. [50]

    Simpler is better: Few-shot semantic segmenta- tion with classifier weight transformer

    Zhihe Lu, Sen He, Xiatian Zhu, Li Zhang, Yi-Zhe Song, and Tao Xiang. Simpler is better: Few-shot semantic segmenta- tion with classifier weight transformer. 2021. 3

    2021

  43. [51]

    Learn to augment: Joint data augmentation and network op- timization for text recognition

    Canjie Luo, Yuanzhi Zhu, Lianwen Jin, and Yongpan Wang. Learn to augment: Joint data augmentation and network op- timization for text recognition. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 13743–13752, 2020. 4 10

    2020

  44. [52]

    The iam-database: an en- glish sentence database for offline handwriting recognition

    Urs-Viktor Marti and Horst Bunke. The iam-database: an en- glish sentence database for offline handwriting recognition. International Journal on Document Analysis and Recogni- tion, 2002. 1, 4

    2002

  45. [53]

    Shuang Mei, Hua Yang, and Z. Yin. An unsupervised- learning-based approach for automated defect inspection on textured surfaces. IEEE Transactions on Instrumentation and Measurement, 2018. 7

    2018

  46. [54]

    Labahn, Tobias Gr ¨uning, and Jochen Z¨ollner

    Johannes Michael, R. Labahn, Tobias Gr ¨uning, and Jochen Z¨ollner. Evaluating sequence-to-sequence models for hand- written text recognition. IEEE International Conference on Document Analysis and Recognition, 2019. 2, 3, 4, 5

    2019

  47. [55]

    Carmon, and Ludwig Schmidt

    John Miller, Rohan Taori, Aditi Raghunathan, Shiori Sagawa, Pang Wei Koh, Vaishaal Shankar, Percy Liang, Y . Carmon, and Ludwig Schmidt. Accuracy on the line: on the strong correlation between out-of-distribution and in- distribution generalization. International Conference on...

    2021

  48. [56]

    Saleh Momeni and B. BabaAli. A transformer-based approach for arabic offline handwritten text recognition. arXiv.org, 2023. 3

    2023

  49. [57]

    Aly Mostafa, Omar Mohamed, Ali Ashraf, Ahmed Elbe- hery, Salma Jamal, Ghada Khoriba, and A. Ghoneim. Oc- former: A transformer-based model for arabic handwritten text recognition. 2021 International Mobile, Intelligent, and Ubiquitous Computing Conference (MIUCC), 2021. 3

    2021

  50. [58]

    Paul, Gagan Madan, Akankshya Mishra, N

    S. Paul, Gagan Madan, Akankshya Mishra, N. Hegde, Pradeep Kumar, and Gaurav Aggarwal. Weakly supervised information extraction from inscrutable handwritten docu- ment images. arXiv, 2023. 3

    2023

  51. [59]

    Kosmos-2: Ground- ing multimodal large language models to the world

    Zhiliang Peng, Wenhui Wang, Li Dong, Yaru Hao, Shaohan Huang, Shuming Ma, and Furu Wei. Kosmos-2: Ground- ing multimodal large language models to the world. ArXiv, abs/2306.14824, 2023. 5, 1

    2023 arXiv

  52. [60]

    Carlos Pe ˜narrubia, J. J. Valero-Mas, and Jorge Calvo- Zaragoza. Self-supervised learning for text recognition: A critical survey. arXiv.org, 2024. 3

    2024

  53. [61]

    Cascianelli, and R

    Vittorio Pippi, S. Cascianelli, and R. Cucchiara. Handwritten text generation from visual archetypes. arXiv.org, 2023. 1, 3, 6

    2023

  54. [62]

    How to Choose Pretrained Handwriting Recognition Models for Single Writer Fine-Tuning, 2023

    Vittorio Pippi, Silvia Cascianelli, Christopher Kermorvant, and Rita Cucchiara. How to Choose Pretrained Handwriting Recognition Models for Single Writer Fine-Tuning, 2023. Version Number: 1. 3, 7

    2023

  55. [63]

    Character-based handwritten text transcription with attention networks.Neural Computing and Applications, 2021

    Jason Poulos and Rafael Valle. Character-based handwritten text transcription with attention networks.Neural Computing and Applications, 2021. 3

    2021

  56. [64]

    Joan Puigcerver. Are multidimensional recurrent layers re- ally necessary for handwritten text recognition? In 14th IAPR International Conference on Document Analysis and Recognition, ICDAR 2017, Kyoto, Japan, November 9-15, 2017, pages 67–72. IEEE, 2017. 1, 3, 4, 5, 2

    2017

  57. [65]

    Learning to learn single domain generalization

    Fengchun Qiao, Long Zhao, and Xi Peng. Learning to learn single domain generalization. Computer Vision and Pattern Recognition, 2020. 2, 3

    2020

  58. [66]

    Toni M. Rath, R. Manmatha, and Victor Lavrenko. A search engine for historical manuscript images. In Proceedings of the 27th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval , page 369–376, New York, NY , USA, 2004. Association for...

    2004

  59. [67]

    Janet C. Read. A study of the usability of handwriting recog- nition for text entry by children.Interacting with Computers, 19(1):57–69, 2007. Moving Face-to-Face communication to Web-based systems. 1

    2007

  60. [68]

    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 Document Analysis Systems , pages 247–259, Cham, 2022. Springer International Publishing. 4

    2022

  61. [69]

    Toselli, V olkmar Frinken, En- rique Vidal, and Josep Llad ´os

    Ver ´onica Romero, Alicia Forn´es, Nicol´as Serrano, Joan An- dreu S ´anchez, Alejandro H. Toselli, V olkmar Frinken, En- rique Vidal, and Josep Llad ´os. The esposalles database: An ancient marriage license corpus for off-line handwriting recognition. Pattern Recognition, 46(...

    2013

  62. [70]

    D. V . Sang and Le Tran Bao Cuong. Improving crnn with efficientnet-like feature extractor and multi-head attention for text recognition. SoICT 2019, 2019. 3

    2019

  63. [71]

    Verzeichniss der Handschriften der Stifts- bibliothek von St

    Gustav Scherrer. Verzeichniss der Handschriften der Stifts- bibliothek von St. Gallen. Halle, 1875. 4

  64. [72]

    The RODRIGO database

    Nicolas Serrano, Francisco Castro, and Alfons Juan. The RODRIGO database. In Proceedings of the Seventh Interna- tional Conference on Language Resources and Evaluation (LREC’10), Valletta, Malta, 2010. European Language Re- sources Association (ELRA). 4

    2010

  65. [73]

    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. In 2019 International Conference on Document Analysis and Recognition (ICDAR), pages 1175–1180, Sydney, Australia,

    2019

  66. [74]

    Spearman

    C. Spearman. General intelligence objectively determined and measured. 1904. 2, 7

    1904

  67. [75]

    Wit: Wikipedia-based image text dataset for multimodal multilingual machine learning

    Krishna Srinivasan, Karthik Raman, Jiecao Chen, Michael Bendersky, and Marc Najork. Wit: Wikipedia-based image text dataset for multimodal multilingual machine learning. In Proceedings of the 44th International ACM SIGIR Confer- ence on Research and Development in Information ...

    2021

  68. [76]

    Toselli, and E

    Joan Andreu S ´anchez, Ver´onica Romero, A. Toselli, and E. Vidal. Icfhr2014 competition on handwritten text recogni- tion on transcriptorium datasets (htrts). 2014 14th Interna- tional Conference on Frontiers in Handwriting Recognition,

    2014

  69. [77]

    Toselli, and Enrique Vidal

    Joan Andreu S ´anchez, Ver ´onica Romero, Alejandro H. Toselli, and Enrique Vidal. Icfhr2016 competition on hand- written text recognition on the read dataset. In 2016 15th In- ternational Conference on Frontiers in Handwriting Recog- nition (ICFHR), pages 630–635, 2016. 4

    2016

  70. [78]

    Toselli, M

    Joan-Andreu S ´anchez, Ver´onica Romero, A. Toselli, M. Vil- legas, and E. Vidal. Icdar2017 competition on handwritten text recognition on the read dataset. 2017 14th IAPR Inter- national Conference on Document Analysis and Recognition (ICDAR), 2017. 1, 2 11

    2017

  71. [79]

    Sun, Tao Kong, Wenchang Zhang, Chao Yang, and Chunfang Liu

    Chuanqi Tan, F. Sun, Tao Kong, Wenchang Zhang, Chao Yang, and Chunfang Liu. A survey on deep transfer learn- ing. International Conference on Artificial Neural Networks,

  72. [80]

    Is it an i or an l: Test- time Adaptation of Text Line Recognition Models, 2023

    Debapriya Tula, Sujoy Paul, Gagan Madan, Peter Garst, Reeve Ingle, and Gaurav Aggarwal. Is it an i or an l: Test- time Adaptation of Text Line Recognition Models, 2023. Version Number: 1. 3

    2023

  73. [81]

    Dhali, and Lambert Schomaker

    Tobias van der Werff, Maruf A. Dhali, and Lambert Schomaker. Writer adaptation for offline text recognition: An exploration of neural network-based methods, 2023. Ver- sion Number: 1. 3

    2023

  74. [82]

    Cascianelli, Nick Michiels, F

    Bram Vanherle, Vittorio Pippi, S. Cascianelli, Nick Michiels, F. Reeth, and R. Cucchiara. Vatr++: Choose your words wisely for handwritten text generation. arXiv.org, 2024. 1, 3, 6

    2024

  75. [83]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neu- ral Information Processing Systems. Curran Associates, Inc.,

  76. [84]

    Generalizing to unseen domains: A survey on domain generalization

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, and Tao Qin. Generalizing to unseen domains: A survey on domain generalization. IEEE Transactions on Knowledge and Data Engineering, 2021. 2, 3

    2021

  77. [85]

    Deep visual domain adaptation: A survey

    Mei Legam Wang and Weihong Deng. Deep visual domain adaptation: A survey. Neurocomputing, 2018. 2, 3

    2018

  78. [86]

    Learning to diversify for single do- main generalization

    Zijian Wang, Yadan Luo, Ruihong Qiu, Zi Huang, and Mahsa Baktashmotlagh. Learning to diversify for single do- main generalization. 2021 IEEE/CVF International Confer- ence on Computer Vision (ICCV), 2021. 2, 3

    2021

  79. [87]

    Weiss, Taghi M

    Karl R. Weiss, Taghi M. Khoshgoftaar, and Dingding Wang. A survey of transfer learning. Journal of Big Data, 2016. 2, 3

    2016

  80. [88]

    Assaying out-of- distribution generalization in transfer learning

    Florian Wenzel, Carl-Johann Simon-Gabriel, David Kern- ert, Bernt Schiele, Andrea Dittadi, Max Horn, Chris Rus- sell, Bernhard Sch ¨olkopf, Peter Gehler, Dominik Zietlow, Thomas Brox, and Francesco Locatello. Assaying out-of- distribution generalization in transfer learning. I...

    2024

  81. [89]

    Rescoring sequence-to-sequence models for text line recog- nition with ctc-prefixes

    Christoph Wick, Jochen Z ¨ollner, and Tobias Gr ¨uning. Rescoring sequence-to-sequence models for text line recog- nition with ctc-prefixes. arXiv: Computer Vision and Pattern Recognition, 2021. 3

    2021

  82. [90]

    Wick, Jochen Z ¨ollner, and Tobias Gr¨uning

    C. Wick, Jochen Z ¨ollner, and Tobias Gr¨uning. Transformer for handwritten text recognition using bidirectional post- decoding. ICDAR, 2021. 3, 4

    2021

  83. [91]

    Garrett Wilson and D. Cook. A survey of unsupervised deep domain adaptation. ACM Transactions on Intelligent Systems and Technology, 2018. 2, 3

    2018

  84. [92]

    Confirmatory factor analysis for applied re- search

    Phil Wood. Confirmatory factor analysis for applied re- search. 2008. 2, 7

    2008

  85. [93]

    Visual anomaly detection for images: A systematic survey

    Jie Yang, Ruijie Xu, Zhiquan Qi, and Yong Shi. Visual anomaly detection for images: A systematic survey. Proce- dia Computer Science, 199:471–478, 2022. The 8th Interna- tional Conference on Information Technology and Quantita- tive Management (ITQM 2020 & 2021): Developing Gl...

    2022

  86. [94]

    Mingkun Yang, Minghui Liao, Pu Lu, Jing Wang, Sheng- gao Zhu, Hualin Luo, Qingzhen Tian, and X. Bai. Read- ing and writing: Discriminative and generative modeling for self-supervised text recognition. ACM Multimedia, 2022. 3

    2022

  87. [95]

    Carbonell, Ruslan Salakhutdinov, and Quoc V

    Zhilin Yang, Zihang Dai, Yiming Yang, Jaime G. Carbonell, Ruslan Salakhutdinov, and Quoc V . Le. Xlnet: General- ized autoregressive pretraining for language understanding. arXiv: Computation and Language, 2019. 3

    2019

  88. [96]

    Sequence-To-Sequence Domain Adaptation Network for Robust Text Image Recognition

    Yaping Zhang, Shuai Nie, Wenju Liu, Xing Xu, Dongxiang Zhang, and Heng Tao Shen. Sequence-To-Sequence Domain Adaptation Network for Robust Text Image Recognition. In 2019 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR) , pages 2735–2744, Long Beach, CA,...

    2019

  89. [97]

    Kaiyang Zhou, Ziwei Liu, Y . Qiao, T. Xiang, and Chen Change Loy. Domain generalization: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence,

  90. [98]

    A comprehensive survey on transfer learning

    Fuzhen Zhuang, Zhiyuan Qi, Keyu Duan, Dongbo Xi, Yongchun Zhu, Hengshu Zhu, Hui Xiong, and Qing He. A comprehensive survey on transfer learning. Proceedings of the IEEE, 2019. 2, 3 12 On the Generalization of Handwritten Text Recognition Models Supplementary Material

    2019

  91. [100]

    Complementary results In this section, we provide additional metrics to the Char- acter Error Rate (CER) presented in the main paper. We provide the following information regarding the results: the main table of the paper (Table 3) is presented in terms of Word Error Rate (WER...

  92. [101]

    Hyperparameters 8.1. Architectures implementation As stated in the main paper, we aimed to follow the imple- mentation closest to the original papers using the available information for those that did not provide code. In all cases, the most significant architectural change oc...

  93. [102]

    Visual and textual divergences In this section, we present the specific numerical metrics for visual and textual divergence across the various domains used in the factor analysis. Prior to presenting these re- sults, we first describe the training procedure for the Con- voluti...

    2016

  94. [103]

    Note that the first k factors span the subspace defined by the first k eigenvectors of the data matrix

    Factor analysis The selection of the number of factors is a crucial criterion for analyzing the outcomes of the factor analysis. Note that the first k factors span the subspace defined by the first k eigenvectors of the data matrix. To determine the number of factors (n), the ...

  95. [2022]

    3, 4, 5, 2

    Series Title: Lecture Notes in Computer Science. 3, 4, 5, 2

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.