Despite having hundreds of millions of speakers, handwritten Devanagari text remains severely underrepresented in publicly available benchmark datasets. Existing resources are limited in scale, focus primarily on isolated characters or short words, and lack controlled lexical content and writer level diversity, which restricts their utility for modern data driven handwriting analysis. As a result, they fail to capture the continuous, fused, and structurally complex nature of Devanagari handwriting, where characters are connected through a shared shirorekha (horizontal headline) and exhibit rich ligature formations. We introduce DohaScript, a large scale, multi writer dataset of handwritten Hindi text collected from 531 unique contributors. The dataset is designed as a parallel stylistic corpus, in which all writers transcribe the same fixed set of six traditional Hindi dohas (couplets). This controlled design enables systematic analysis of writer specific variation independent of linguistic content, and supports tasks such as handwriting recognition, writer identification, style analysis, and generative modeling. The dataset is accompanied by non identifiable demographic metadata, rigorous quality curation based on objective sharpness and resolution criteria, and page level layout difficulty annotations that facilitate stratified benchmarking. Baseline experiments demonstrate clear quality separation and strong generalization to unseen writers, highlighting the dataset's reliability and practical value. DohaScript is intended to serve as a standardized and reproducible benchmark for advancing research on continuous handwritten Devanagari text in low resource script settings.
Figures & tables
Dataset
Script(s)
Page-level Samples (# Pages)
# Writers
Text Type
Demographics
IIIT-HW-Words ( Gongidi and Jawahar, 2021 )
Indic (multi)
– (word-only dataset)
135
Isolated words
No
IIIT-HW-Dev ( Dutta et al., 2018 )
Hindi (Devanagari)
– (word-only dataset)
12
Continuous words
Limited
Numerals/Vowels Dataset ( Prashanth et al., 2022 )
Devanagari
– (character-only dataset)
∼ 3800 subjects
Isolated characters
Yes
CHIPS (PLATTER) ( Kasuba et al., 2025 )
Indic (multi)
25,290 pages
–
Continuous
No
Parimal Hindi ( Dey et al., 2025 )
Hindi (Devanagari)
500 pages
10
Continuous paragraphs
Yes
AnciDev ( Sharma et al., 2025 )
Ancient Devanagari
500 pages
–
Continuous lines
No
\fontspec_if_language:nTF ENG \addfontfeature Language=English Table 1 . Quantitative comparison of existing Indic handwriting datasets with the proposed dataset.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 1 . Text corpus comprising six dohas used in the study.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 2 . Comprehensive coverage of Devanagari characters in the selected dohas , including consonants by articulation category, vowel classes, vowel diacritics (mātrās), halant, and common conjuncts.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 3 . Age distribution of the 531 participants in the dataset. The dashed line indicates the median age. The detailed distribution is shown in Table \fontspec_if_language:nTFENG\addfontfeatureLanguage=English8 in the Appendix \fontspec_if_language:nTFENG\addfontfeatureLanguage=EnglishB .
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 4 . Frequency distribution of handwriting samples across states of India. The tabulated state wise distribution is show in Table \fontspec_if_language:nTFENG\addfontfeatureLanguage=English7 in Appendix \fontspec_if_language:nTFENG\addfontfeatureLanguage=EnglishA .
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 5 . Sample images from the dataset. Each segment corresponds to one handwritten page instance.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 6 . Blur score distribution with initial binary threshold labeling (before CNN refinement).
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 7 . Blur score distribution by quality class. Boxes show inter quartile ranges with medians (horizontal lines) and means (green triangles). Outliers appear as circles. Class separation validates stratification, while MediumGood overlap explains reduced classification performance in boundary regions.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 8 . CNN architecture for handwriting quality classification. Three convolutional blocks extract hierarchical features, followed by adaptive pooling and fully connected layers for quality prediction.
\fontspec_if_language:nTF ENG \addfontfeature Language=English Table 3 . Per-class performance on test set (107 samples)
\fontspec_if_language:nTF ENG \addfontfeature Language=English Figure 9 . Representative handwritten pages from each segmentation difficulty class. Easy samples exhibit clear inter-line separation, Medium pages contain moderate spacing variation, and Complex pages show overlap and irregular baselines leading to ambiguous line boundaries.
Metric
Value
Total Documents
531
Perfect Segmentation (12/12 lines)
157 (29.57%)
Mean Segmentation Score
43.46 ± 22.55
95% CI
[41.54, 45.39]
Median Score
42.91
Mean Lines Detected
11.22 ± 2.92
\fontspec_if_language:nTF ENG \addfontfeature Language=English Table 4 . Line segmentation performance on DohaScript (N=531).
Difficulty
Count
Percentage
Score Range
Easy
110
20.7%
65.1–91.8
Medium
141
26.6%
45.9–66.4
Complex
280
52.7%
7.6–46.7
\fontspec_if_language:nTF ENG \addfontfeature Language=English Table 5 . Distribution of document segmentation difficulty.
Error (lines)
Count
Percentage
0
157
29.6%
1
138
26.0%
2
69
13.0%
3
64
12.1%
4
28
5.3%
5
28
5.3%
\fontspec_if_language:nTF ENG \addfontfeature Language=English Table 6 . Distribution of line detection errors.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
S.No.
State
Participants
1
Andhra Pradesh
15
2
Arunachal Pradesh
1
3
Assam
3
4
Bihar
150
5
Chandigarh
2
6
Chhattisgarh
3
Appendix
\fontspec_if_language:nTF ENG \addfontfeature Language=EnglishTable 7. State-wise distribution of participants.
Age
Participants
Age
Participants
7
4
19
99
8
4
20
76
9
14
21
70
10
2
22
47
11
7
23
6
12
9
24
3
Appendix
\fontspec_if_language:nTF ENG \addfontfeature Language=EnglishTable 8. Distribution of participants across different age groups.
In this work, we target Handwritten Text Recognition (HTR) in low-resource scenarios, which arise from underrepresented languages, rare scripts, and degraded visual conditions typical of historical documents. We introduce SCAM (Sahidic Coptic Ancient Manuscripts), a new line-level dataset built from digitized ancient manuscripts written in the extinct Sahidic Coptic dialect. The dataset reflects a realistic and challenging setting, as it combines heterogeneous acquisition conditions across libraries with typical manuscript degradations such as ink fading, bleed-through, and material deterioration. In addition to visual complexity, SCAM poses significant linguistic challenges due to the scarcity of resources for Sahidic Coptic, its uncommon alphabet, and dialect-specific diacritics. To support research in low-resource HTR, we benchmark several state-of-the-art approaches based on different paradigms, highlighting their limitations and strengths in this setting. Our results underline the gap between current HTR performance on well-resourced modern scripts and historically grounded, low-resource scenarios, thus providing a reference point for future developments.
To train handwritten text recognition systems we need word images and their corresponding transcriptions, and these transcriptions are produced manually. For a script that can be read by only a small number of specialists, this manual transcription is a limitation, because the trained models are supposed to save the time of those same specialists. A relevant question therefore arises: how many transcriptions are needed before a recogniser becomes useful, and how much of that cost can pretraining remove? In this study the answer is measured directly for handwritten Devanagari. We keep the recogniser, optimiser and evaluation protocol the same and change only the number of real transcribed words used for fine-tuning across nine budgets from 10 to 4,000 and four initialisation regimes, with six seeds at every point. The resulting curves are then converted into annotation-equivalent terms. A CER of 0.50 is reached by supervised synthetic pretraining using only 81 transcribed words, whereas random initialisation requires 355, which gives a label multiplier of 4.40 [3.56, 4.99]. There is a zero-shot reference point as well: with no real transcribed words at all, this pretraining is worth about 136 of them. This advantage gets smaller as the target accuracy improves, and at the most demanding target we measure, it cannot be distinguished from no saving at all. A fourth arm in which only the encoder is transferred separates the effect of the pretraining method from that of transfer scope, and masked image modelling is observed to transfer negatively over a bounded range of budgets. We emphasise that the scarcity in this study is constructed by subsampling a large corpus.
Manglesh Kumar Pandey, Sumit Kumar Banshal
Alliance School of Liberal Arts and Sciences, Alliance University Bengaluru, India · Department of Computer Science and Engineering, Alliance University Bengaluru, India
We introduce the Burmese Handwritten Digit Dataset (BHDD), a collection of 87,561 grayscale images of handwritten Burmese digits in ten classes. Each image is 28x28 pixels, following the MNIST format. The training set has 60,000 samples split evenly across classes; the test set has 27,561 samples with class frequencies as they arose during collection. Over 150 people of different ages and backgrounds contributed samples. We analyze the dataset's class distribution, pixel statistics, and morphological variation, and identify digit pairs that are easily confused due to the round shapes of the Myanmar script. Simple baselines (an MLP, a two-layer CNN, and an improved CNN with batch normalization and augmentation) reach 99.40%, 99.75%, and 99.83% test accuracy respectively. BHDD is available under CC BY-SA 4.0 at https://github.com/baseresearch/BHDD