pith. machine review for the scientific record. sign in

arxiv: 2604.22515 · v2 · submitted 2026-04-24 · 💻 cs.CV · cs.LG

Recognition: unknown

Different Strokes for Different Folks: Writer Identification for Historical Arabic Manuscripts

Ariel Justine N. Panopio, Hamza A. Abushahla, Layth Al-Khairulla, Mohamed I. AlHajri

Pith reviewed 2026-05-08 12:28 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords writer identificationhistorical Arabic manuscriptsMuharaf datasetconvolutional neural networksattention mechanismspage-disjoint protocolhandwriting analysiscultural heritage documents
0
0 comments X

The pith

Fine-tuned DenseNet201 with attention reaches 99.05% top-1 accuracy identifying writers from individual lines in historical Arabic manuscripts.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper manually verifies and expands writer labels in the Muharaf dataset from 6,858 to 21,249 lines, retaining 18,987 after filtering. It trains and benchmarks convolutional neural network models that incorporate attention mechanisms to perform closed-set writer identification from line images, including lines written by two people treated as composite classes. Under random splits of individual lines the best model records 99.05% top-1 accuracy, 99.73% top-5 accuracy and 97.44% F1-score. When entire pages are kept disjoint between training and test sets the same architecture drops to 78.61% top-1 accuracy, showing how much performance depends on page-level cues rather than handwriting style alone. The expanded labels and dual-protocol baselines supply historians and linguists with a concrete resource for provenance and authenticity work on culturally significant documents.

Core claim

By expanding the labeled subset of the Muharaf dataset of historical Arabic manuscripts and training a CNN with attention, the authors establish the first reported baselines for closed-set writer identification under both line-level and page-disjoint protocols. A fine-tuned DenseNet201 model with attention achieves 99.05% Top-1 accuracy, 99.73% Top-5 accuracy and 97.44% F1-score when lines are split randomly. The same model records 78.61% Top-1, 87.79% Top-5 and 66.55% F1-score under the stricter page-disjoint protocol, thereby quantifying the contribution of page context.

What carries the argument

A convolutional neural network with attention layers, using DenseNet201 as the feature extractor and trained on line images for closed-set classification by writer identity, with fourteen configurations and ablations across feature extractors and training regimes.

If this is right

  • Historians obtain a practical method for attributing authorship in unsigned Arabic manuscripts once the model is applied to new line images.
  • The performance gap between protocols directly measures how much page layout and context contribute to identification accuracy.
  • Lines written by two different hands can be handled by treating each writer pair as a distinct composite class.
  • The corrected and expanded labeled subset of 18,987 lines becomes a reusable training resource for further digital-heritage work.
  • Benchmark numbers under both protocols serve as a reference point for comparing future writer-identification techniques.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same labeling-plus-attention pipeline could be repeated on manuscripts in other scripts to create comparable public datasets.
  • Future models might improve by explicitly disentangling handwriting strokes from page layout features to close the page-disjoint gap.
  • Automated consistency checks on writer labels could lower the cost of scaling the approach to much larger manuscript collections.

Load-bearing premise

The manually verified and expanded writer labels from the original partial annotations contain no systematic inconsistencies that affect model training or evaluation.

What would settle it

A second independent manual review of a random subset of the expanded line labels, followed by retraining the DenseNet201 model and measuring any drop in the reported top-1 and F1-score figures.

read the original abstract

Handwritten Arabic manuscripts preserve the Arab world's intellectual and cultural heritage, and writer identification supports provenance, authenticity verification, and historical analysis. Using the Muharaf dataset of historical Arabic manuscripts, we evaluate writer identification from individual line images and, to the best of our knowledge, provide the first baselines reported under both line-level and page-disjoint evaluation protocols. Since the dataset is only partially labeled for writer identification, we manually verified and expanded writer labels in the public portion from 6,858 (28.00%) to 21,249 lines (86.75%) out of 24,495 line images, correcting inconsistencies and removing non-handwritten text. After further filtering, we retained 18,987 lines (77.51%). We propose a Convolutional Neural Network (CNN)-based model with attention mechanisms for closed-set writer identification, including rare two-writer lines modeled as composite writer-pair classes. We benchmark fourteen configurations and conduct ablations across different feature extractors and training regimes. To assess generalization to unseen pages, the page-disjoint protocol assigns all lines from each page to a single split. Under the line-level protocol, a fine-tuned DenseNet201 with attention achieves 99.05% Top-1 accuracy, 99.73% Top-5 accuracy, and 97.44% F1-score. Under the more challenging page-disjoint protocol, the best observed results are 78.61% Top-1 accuracy, 87.79% Top-5 accuracy, and 66.55% F1-score, thus quantifying the impact of page-level cues. By expanding the Muharaf dataset's labeled subset and reporting both protocols, we provide a clearer benchmark and a practical resource for historians and linguists engaged with culturally and historically significant documents. The code and implementation details are available on GitHub.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper expands the labeled subset of the Muharaf dataset of historical Arabic manuscripts by manually verifying and correcting writer labels, increasing coverage from 6,858 to 21,249 lines (retaining 18,987 after filtering). It introduces a CNN architecture with attention for closed-set writer identification (treating rare two-writer lines as composite classes), benchmarks 14 configurations across feature extractors under both line-level and page-disjoint protocols, and reports peak results of 99.05% Top-1 / 99.73% Top-5 accuracy (line-level) and 78.61% Top-1 / 87.79% Top-5 accuracy (page-disjoint) with a fine-tuned DenseNet201+attention model. Code is released on GitHub.

Significance. If the expanded labels prove reliable, the work supplies the first reported baselines for writer identification on this culturally significant dataset under two protocols that explicitly quantify page-level cue leakage. The ablations across multiple extractors, the dual-protocol design, and the public code release are concrete strengths that enable reproducibility and support downstream use by historians and digital-humanities researchers.

major comments (1)
  1. [Dataset preparation] Dataset preparation section (description of label expansion and filtering): The manual verification process that expands writer labels from 6,858 to 21,249 lines (and retains 18,987 after filtering) is presented without reporting the number of annotators, inter-annotator agreement, sampling strategy for verification, or concrete examples of corrected inconsistencies. This is load-bearing for the central empirical claims because the headline accuracies (99.05% Top-1 line-level; 78.61% Top-1 page-disjoint) are only interpretable if the labels are free of systematic noise; historical Arabic hands are known to be visually confusable, so any unquantified labeling error would directly bias both evaluation protocols.
minor comments (2)
  1. [Abstract / Results] Abstract and results tables: The exact breakdown of the 5,262 lines removed during filtering (non-handwritten text, inconsistencies, etc.) is not tabulated; adding a small supplementary table would improve transparency without altering the narrative.
  2. [Methodology] Methodology: The attention mechanism is described at a high level; a brief diagram or equation showing how the attention weights are computed and fused with the DenseNet features would aid reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of our work's significance and for the constructive feedback on dataset preparation. We address the major comment below and commit to revisions that strengthen the manuscript without altering its core claims.

read point-by-point responses

  1. Referee: [Dataset preparation] Dataset preparation section (description of label expansion and filtering): The manual verification process that expands writer labels from 6,858 to 21,249 lines (and retains 18,987 after filtering) is presented without reporting the number of annotators, inter-annotator agreement, sampling strategy for verification, or concrete examples of corrected inconsistencies. This is load-bearing for the central empirical claims because the headline accuracies (99.05% Top-1 line-level; 78.61% Top-1 page-disjoint) are only interpretable if the labels are free of systematic noise; historical Arabic hands are known to be visually confusable, so any unquantified labeling error would directly bias both evaluation protocols.

    Authors: We agree that additional transparency on the label verification process is warranted to support interpretability of the results. In the revised manuscript we will expand the Dataset preparation section with the following: the verification and correction of existing labels was performed manually by the authors (who have relevant expertise in historical Arabic manuscripts); the process was exhaustive rather than sampled (all lines in the public portion were reviewed to expand coverage, correct writer attributions, and remove non-handwritten elements); and we will include 2-3 concrete examples of corrected inconsistencies (e.g., swapped writer IDs arising from visually similar hands). We did not compute formal inter-annotator agreement because the task was verification/correction of pre-existing labels rather than independent multi-annotator annotation; we will explicitly note this as a limitation. These changes will directly address the referee's concern while preserving the reported experimental outcomes. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical benchmark on held-out splits

full rationale

The paper reports standard CNN training and evaluation results (DenseNet201 with attention, etc.) on manually expanded Muharaf labels under line-level and page-disjoint protocols. No equations, derivations, or first-principles claims exist that reduce reported accuracies to fitted parameters or self-defined quantities within the paper. Label expansion is described as a manual verification step with no self-referential fitting or prediction loop. The central results are direct empirical measurements on data splits, with no load-bearing self-citations or ansatz smuggling. This is a self-contained benchmark paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central performance claims rest on the correctness of the manually expanded writer labels and on the assumption that line images are sufficiently independent once page-disjoint splits are enforced. No new physical or mathematical entities are introduced.

pith-pipeline@v0.9.0 · 5659 in / 1300 out tokens · 28302 ms · 2026-05-08T12:28:59.955534+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 7 canonical work pages · 2 internal anchors

  1. [1]

    Journal of African Resilience and Advancement Research (2024)

    Boyi, A.M., Yusuf, M.B., Isa, M.M.: The evolution and history of the arabic language. Journal of African Resilience and Advancement Research (2024)

    2024

  2. [2]

    International Journal on Document Analysis and Recog- nition (IJDAR),20, 173 (2017)

    Asi, A., Abdalhaleem, A., Fecker, D., M¨argner, V., et al.: On writer identification for arabic historical manuscripts. International Journal on Document Analysis and Recog- nition (IJDAR),20, 173 (2017)

    2017

  3. [3]

    Multimedia Tools and Applications,78, 10889 (2019)

    Rehman, A., Naz, S., Razzak, M.I.: Writer identifica- tion using machine learning approaches: a comprehen- sive review. Multimedia Tools and Applications,78, 10889 (2019)

    2019

  4. [4]

    In: 2020 19th IEEE International Conference on Machine Learning and Ap- plications (ICMLA) (IEEE, 2020), pp

    Chammas, M., Makhoul, A., Demerjian, J.: Writer iden- tification for historical handwritten documents using a single feature extraction method. In: 2020 19th IEEE International Conference on Machine Learning and Ap- plications (ICMLA) (IEEE, 2020), pp. 1–6

    2020

  5. [5]

    Multimedia Tools and Applications,83(18), 54569 (2024)

    Chammas, M., Makhoul, A., Demerjian, J., Dannaoui, E.: An end-to-end deep learning system for writer identi- fication in handwritten arabic manuscripts. Multimedia Tools and Applications,83(18), 54569 (2024)

    2024

  6. [6]

    In: International Confer- ence on Document Analysis and Recognition (Springer, 2021), pp

    Ngo, T.T., Nguyen, H.T., Nakagawa, M.: A-vlad: An end- to-end attention-based neural network for writer identifi- cation in historical documents. In: International Confer- ence on Document Analysis and Recognition (Springer, 2021), pp. 396–409

    2021

  7. [7]

    Advances in Neural Information Processing Systems,37, 58525 (2024)

    Saeed, M., Chan, A., Mijar, A., Habchi, G., et al.: Muharaf: Manuscripts of handwritten arabic dataset for cursive text recognition. Advances in Neural Information Processing Systems,37, 58525 (2024)

    2024

  8. [8]

    In: 2017 1st International workshop on arabic script analysis and recognition (ASAR) (IEEE, 2017), pp

    Abdelhaleem, A., Droby, A., Asi, A., Kassis, M., et al.: Wahd: a database for writer identification of arabic his- torical documents. In: 2017 1st International workshop on arabic script analysis and recognition (ASAR) (IEEE, 2017), pp. 64–68

    2017

  9. [9]

    Pattern Recognition,47(3), 1096 (2014)

    Mahmoud, S.A., Ahmad, I., Al-Khatib, W.G., Alshayeb, M., et al.: Khatt: An open arabic offline handwritten text database. Pattern Recognition,47(3), 1096 (2014)

    2014

  10. [10]

    arXiv preprint arXiv:2410.02179 (2024)

    Chan, A., Mijar, A., Saeed, M., Wong, C.W., et al.: Hat- former: Historic handwritten arabic text recognition with transformers. arXiv preprint arXiv:2410.02179 (2024)

    work page arXiv 2024

  11. [11]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2016), pp

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2016), pp. 770–778

    2016

  12. [12]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014)

    work page internal anchor Pith review arXiv 2014

  13. [13]

    Pattern Recognition Letters,129, 137 (2020)

    Cilia, N.D., De Stefano, C., Fontanella, F., Marrocco, C., et al.: An end-to-end deep learning system for medieval writer identification. Pattern Recognition Letters,129, 137 (2020)

    2020

  14. [14]

    Journal of Imaging,6(9), 89 (2020)

    Cilia, N.D., De Stefano, C., Fontanella, F., Marrocco, C., et al.: An experimental comparison between deep learning and classical machine learning approaches for writer identification in medieval documents. Journal of Imaging,6(9), 89 (2020)

    2020

  15. [15]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2016), pp

    Arandjelovic, R., Gronat, P., Torii, A., Pajdla, T., et al.: Netvlad: Cnn architecture for weakly supervised place recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2016), pp. 5297–5307

    2016

  16. [16]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp

    Zhang, H., Xue, J., Dana, K.: Deep ten: Texture encoding network. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp. 708– 717

    2017

  17. [17]

    In: Asian Conference on Pattern Recognition (Springer, 2021), pp

    Srivastava, A., Chanda, S., Pal, U.: Exploiting multi-scale fusion, spatial attention and patch interaction techniques for text-independent writer identification. In: Asian Conference on Pattern Recognition (Springer, 2021), pp. 203–217

    2021

  18. [18]

    In: 2016 15th international conference on frontiers in handwriting recognition (ICFHR) (IEEE, 2016), pp

    Xing, L., Qiao, Y.: Deepwriter: A multi-stream deep cnn for text-independent writer identification. In: 2016 15th international conference on frontiers in handwriting recognition (ICFHR) (IEEE, 2016), pp. 584–589

    2016

  19. [19]

    Pattern Analysis and Applications,26(3), 1539 (2023)

    Chahi, A., El-Merabet, Y., Ruichek, Y., Touahni, R.: An effective deepwinet cnn model for off-line text- independent writer identification. Pattern Analysis and Applications,26(3), 1539 (2023)

    2023

  20. [20]

    Pattern Recognition,63, 451 (2017)

    He, S., Schomaker, L.: Writer identification using curvature-free features. Pattern Recognition,63, 451 (2017)

    2017

  21. [21]

    In: Proc

    Pechwitz, M., Maddouri, S.S., M ¨argner, V., Ellouze, N., et al.: Ifn/enit-database of handwritten arabic words. In: Proc. of CIFED, vol. 2 (Citeseer, 2002), vol. 2, pp. 127–136

    2002

  22. [22]

    Advances in neural information processing systems,30(2017)

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., et al.: Attention is all you need. Advances in neural information processing systems,30(2017)

    2017

  23. [23]

    In- ternational journal on document analysis and recognition, 5, 39 (2002)

    Marti, U.V., Bunke, H.: The iam-database: an english sentence database for offline handwriting recognition. In- ternational journal on document analysis and recognition, 5, 39 (2002)

    2002

  24. [24]

    In: 2013 12th international conference on document analysis and recognition (IEEE, 2013), pp

    Kleber, F., Fiel, S., Diem, M., Sablatnig, R.: Cvl-database: An off-line database for writer retrieval, writer identifi- cation and word spotting. In: 2013 12th international conference on document analysis and recognition (IEEE, 2013), pp. 560–564 19

    2013

  25. [25]

    Pattern Recognition,48(12), 4036 (2015)

    He, S., Wiering, M., Schomaker, L.: Junction detection in handwritten documents and its application to writer identification. Pattern Recognition,48(12), 4036 (2015)

    2015

  26. [26]

    arXiv preprint arXiv:2404.07602 (2024)

    Kumar, V., Sundaram, S.: Attention based end to end network for offline writer identification on word level data. arXiv preprint arXiv:2404.07602 (2024)

    work page arXiv 2024

  27. [27]

    IEEE Access (2025)

    Okawa, M.: Multistage convolutional neural network with deformable attention for word-level offline text- independent writer identification. IEEE Access (2025)

    2025

  28. [28]

    In: IFIP Interna- tional Conference on Artificial Intelligence Applications and Innovations (Springer, 2025), pp

    Majithia, A., Pedersen, A.P., Grossberg, M.: An inte- grated convolutional and transformer architecture for word-based handwriter identification. In: IFIP Interna- tional Conference on Artificial Intelligence Applications and Innovations (Springer, 2025), pp. 44–59

    2025

  29. [29]

    In: International Workshop on Document Analysis Systems (Springer, 2022), pp

    Koepf, M., Kleber, F., Sablatnig, R.: Writer identification and writer retrieval using vision transformer for forensic documents. In: International Workshop on Document Analysis Systems (Springer, 2022), pp. 352–366

    2022

  30. [30]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A.: An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    work page internal anchor Pith review arXiv 2010

  31. [31]

    In: 2013 12th International conference on document analysis and recognition (IEEE, 2013), pp

    Louloudis, G., Gatos, B., Stamatopoulos, N., Papandreou, A.: Icdar 2013 competition on writer identification. In: 2013 12th International conference on document analysis and recognition (IEEE, 2013), pp. 1397–1401

    2013

  32. [32]

    fatnassi et al

    Fatnassi, I., Khamekhem Jemni, S., Ammar, S., Kessen- tini, Y.: St-wid: Self-supervised transformer for writer identification in arabic handwritten scripts: I. fatnassi et al. Signal, Image and Video Processing,19(14), 1190 (2025)

    2025

  33. [33]

    In: 2012 international con- ference on frontiers in handwriting recognition (IEEE, 2012), pp

    Mezghani, A., Kanoun, S., Khemakhem, M., El Abed, H.: A database for arabic handwritten text image recognition and writer identification. In: 2012 international con- ference on frontiers in handwriting recognition (IEEE, 2012), pp. 399–402

    2012

  34. [34]

    International Journal on Document Analysis and Recognition (IJDAR),27(4), 529 (2024)

    Briber, A., Chibani, Y.: Open writer identification from handwritten text fragments using lite convolutional neural network. International Journal on Document Analysis and Recognition (IJDAR),27(4), 529 (2024)

    2024

  35. [35]

    Expert Systems with Applications, 242, 122704 (2024)

    Yang, J., Shokouhifar, M., Yee, L., Khan, A.A., et al.: Dt2f-tlnet: A novel text-independent writer identification and verification model using a combination of deep type- 2 fuzzy architecture and transfer learning networks based on handwriting data. Expert Systems with Applications, 242, 122704 (2024)

    2024

  36. [36]

    ACM Transactions on Asian and Low-Resource Language Information Pro- cessing,24(7), 1 (2025)

    Maitra, A., Mitra, S., Manna, S., Bhattacharya, S., et al.: Decorrelation-based self-supervised visual representa- tion learning for writer identification. ACM Transactions on Asian and Low-Resource Language Information Pro- cessing,24(7), 1 (2025)

    2025

  37. [37]

    Data in Brief,41, 107947 (2022)

    Khan, M.A.: Arabic handwritten alphabets, words and paragraphs per user (ahawp) dataset. Data in Brief,41, 107947 (2022)

    2022

  38. [38]

    In: International Conference on Cyber- security and Artificial Intelligence Strategies (Springer, 2025), pp

    Khalaif, M.Z., Younis, M.T.: An enhanced deep learning approach for writer identification and verification using corner detection. In: International Conference on Cyber- security and Artificial Intelligence Strategies (Springer, 2025), pp. 232–245

    2025

  39. [39]

    Journal of Forensic Sciences,47(4), 856 (2002)

    Srihari, S.N., Cha, S.H., Arora, H., Lee, S.: Individuality of handwriting. Journal of Forensic Sciences,47(4), 856 (2002)

    2002

  40. [40]

    Pattern Recognition Letters,34(10), 1196 (2013)

    Djeddi, C., Siddiqi, I., Souici-Meslati, L., Ennaji, A.: Text-independent writer recognition using multi-script handwritten texts. Pattern Recognition Letters,34(10), 1196 (2013)

    2013

  41. [41]

    In: 2016 23rd International Conference on Pattern Recognition (ICPR) (IEEE, 2016), pp

    Bertolini, D., Oliveira, L.S., Sabourin, R.: Multi-script writer identification using dissimilarity. In: 2016 23rd International Conference on Pattern Recognition (ICPR) (IEEE, 2016), pp. 3025–3030

    2016

  42. [42]

    Proceedings of the Soviet physics doklady (1966)

    Levenshtein, V.: Binary codes capable of correcting deletions, insertions, and reversals. Proceedings of the Soviet physics doklady (1966)

    1966

  43. [43]

    In: 2009 IEEE conference on computer vision and pattern recognition (Ieee, 2009), pp

    Deng, J., Dong, W., Socher, R., Li, L.J., et al.: Imagenet: A large-scale hierarchical image database. In: 2009 IEEE conference on computer vision and pattern recognition (Ieee, 2009), pp. 248–255

    2009

  44. [44]

    IEEE transactions on pattern analysis and machine intelligence,37(9), 1904 (2015)

    He, K., Zhang, X., Ren, S., Sun, J.: Spatial pyramid pooling in deep convolutional networks for visual recog- nition. IEEE transactions on pattern analysis and machine intelligence,37(9), 1904 (2015)

    1904

  45. [45]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp

    Huang, G., Liu, Z., Van Der Maaten, L., Weinberger, K.Q.: Densely connected convolutional networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp. 4700–4708

    2017

  46. [46]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp

    Chollet, F.: Xception: Deep learning with depthwise separable convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition (2017), pp. 1251–1258

    2017

  47. [47]

    In: 2019 IEEE/CVF Interna- tional Conference on Computer Vision (ICCV) (2019), pp

    Howard, A., Sandler, M., Chen, B., Wang, W., et al.: Searching for mobilenetv3. In: 2019 IEEE/CVF Interna- tional Conference on Computer Vision (ICCV) (2019), pp. 1314–1324. DOI 10.1109/ICCV.2019.00140

    work page doi:10.1109/iccv.2019.00140 2019

  48. [48]

    In: Proceedings of the IEEE conference on com- puter vision and pattern recognition (2018), pp

    Hu, J., Shen, L., Sun, G.: Squeeze-and-excitation net- works. In: Proceedings of the IEEE conference on com- puter vision and pattern recognition (2018), pp. 7132– 7141

    2018

  49. [49]

    arXiv preprint arXiv:2508.15008 (2025)

    Abushahla, H.A., Varam, D., Panopio, A.J.N., AlHajri, M.I.: Neural network quantization for microcontrollers: A comprehensive survey of methods, platforms, and applications. arXiv preprint arXiv:2508.15008 (2025)

    work page arXiv 2025

  50. [50]

    Youssef Bey Al-Helou,

    Mayr, M., Dreier, M., Kordon, F., Seuret, M., et al.: Zero- shot paragraph-level handwriting imitation with latent diffusion models. International Journal of Computer Vision, pp. 1–22 (2025) 20 Appendix A Dataset Distribution Under Protocol A 0 200 400 600 800 Number of Lines Father Elias Shibli Barakat witnesses Nehme Elias Mikhail Maronite children of A...

    work page arXiv 2025