Pith. sign in

REVIEW 3 major objections 5 minor 90 references

Bengali scene text recognition fails primarily through visual mis-recognition, not orthographic confusion, according to a new 2,535-image in-the-wild benchmark.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

BANGLAWILD is the first in-the-wild Bengali scene text benchmark with dual verbatim/standard labels, and its evaluation shows visual mis-recognition dominates errors while conjunct-related errors are nearly closed.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Solid benchmark, but the visual-dominance claim rests on a string-only classifier that has not been validated; the dataset is the real contribution. the 3 major comments →

arxiv 2608.03884 v1 pith:GCBLWQKO submitted 2026-08-04 cs.CV cs.CL

BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models

Sadab Shiper , Tawsif Tashwar Dipto , Mir Md Inzamam , Eshat Tanzeem This is my paper
classification cs.CV cs.CL
keywords Bengali scene text recognitionin-the-wild benchmarkvision-language modelserror taxonomyvisual mis-recognitionorthographic confusionconjunct/hasanta errorsLoRA fine-tuning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

BanglaWild is a first attempt to measure how well modern systems actually read Bengali text in the wild: 2,535 mobile-phone photographs of walls, signboards, vehicles, packaging, and banners, each with a verbatim gold transcription and an orthographically standard form where the in-image spelling deviates. The paper's central claim is that on this benchmark, visual mis-recognition—not orthographic confusion—dominates failure, with visual categories making up roughly 56–60% of errors for the strongest systems and conjunct/hasanta errors below 2%, a profile that also holds for the one conventional OCR baseline that reads Bengali reliably. It also finds that larger models do not consistently beat smaller ones within the same family, that prompt language mainly gates cross-script drift, and that LoRA fine-tuning rescues catastrophic failures in weak models without lifting the ceiling on already competent ones. If these findings hold, they redirect Bengali OCR research from conjunct decoding toward visual robustness, and they argue for dual verbatim/standard evaluation whenever generative models transcribe text with non-standard spelling.

Core claim

The paper introduces BANGLAWILD, a benchmark of 2,535 in-the-wild Bengali scene text images paired with verbatim gold transcriptions and, for the 3.43% of images with non-standard spelling, an orthographically standard form. Using a fifteen-class deterministic error taxonomy applied to four VLM families and three conventional OCR baselines, it reports that lexical substitution, single-character visual misreads, and multi-character word corruption together account for ~56–60% of Gemini-2.5-Flash and Qwen3.5-9B errors, while conjunct/hasanta errors contribute 1.0–1.6%; EasyOCR shows the same visual-dominant profile at a 3.7× visual-to-orthographic ratio. The paper interprets this as inverting

What carries the argument

The load-bearing object is the BANGLAWILD benchmark itself, with its dual verbatim/standard gold transcription: the verbatim form makes recognition errors measurable while the standard form exposes silent over-correction. The argument is carried by a fifteen-class deterministic error taxonomy (Appendix H) that classifies every (ground truth, prediction) pair by a rule-based cascade and edit-distance thresholds, producing the visual-versus-orthographic split; the classifier sees only the string pair, never the image. Supporting mechanisms are the LLM-as-a-Judge protocol (validated against native annotators, ρ=0.89) that supplies a meaning-aware score alongside CER/WER, and LoRA fine-tuning us

Load-bearing premise

The claim that ~60% of errors are visual rests on a rule-based classifier that judges errors from the text pair alone, without seeing the image; if its edit-distance categories do not track the true perceptual cause, the visual-dominance finding would be an artifact.

What would settle it

Take a sample of mis-transcriptions and have native Bengali readers label the cause (visual confusion vs spelling choice) from the image; compare with the taxonomy's visual/orthographic assignment. If human labels disagree with the classifier's visual majority, the paper's central finding is an artifact of edit-distance heuristics.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Bengali OCR research should de-emphasize conjunct decoding: conjunct/hasanta errors are 1.0–1.6% of VLM errors and 4.6% for EasyOCR, far below the ~56–60% share of visual mis-recognition categories.
  • Benchmarks need dual verbatim/standard annotations: over-correction and fluent hallucination are structurally invisible to conventional OCR and to edit-distance metrics, yet they dominate deployment risk for addresses, receipts, and medical packaging.
  • Prompt language has a narrow, script-gating effect: Bengali prompts reduce cross-script contamination (e.g., 6.8%→5.5% on Gemini) but leave visual error categories unchanged within ±1.5 pp.
  • LoRA fine-tuning on 1,268 images converts catastrophic failures (refusals, off-target captioning) into ordinary errors but slightly regresses the strongest open model (Qwen3.5-9B: 20.38→21.53 CER), so small-scale adaptation does not lift the ceiling.
  • Rankings depend on the metric: CER and an LLM judge diverge at the top, so deployment decisions should look at both, and tokenization/Unicode artifacts inflate WER by 22–25% of errors without changing readability.

Where Pith is reading between the lines

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

  • If visual mis-recognition is the true bottleneck, then accuracy should improve more from better detection and image preprocessing (e.g., curved-text rectification, clutter reduction) than from stronger language priors; this is testable by re-running the benchmark with rectified images.
  • The non-monotonic scaling across model sizes suggests training-data composition or benchmark overlap effects; a controlled study holding architecture fixed and varying only Bengali scene-text pretraining data could settle it.
  • The dual verbatim/standard annotation could be adopted for other low-resource scripts with non-standard spelling, turning silent over-correction into a measurable quantity rather than an anecdote.
  • Because prompt language only gates cross-script contamination, prompting in Bengali is a cheap but limited fix; the paper's profile predicts that visual robustness gains will require model-level changes, not instruction changes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces BANGLAWILD, a benchmark of 2,535 in-the-wild Bengali scene text images with verbatim gold transcriptions, categorical and diagnostic attribute labels, and dual verbatim/standard-form annotation. It evaluates 15 vision-language models, 3 conventional OCR systems, and 6 LoRA fine-tuned variants under three prompting strategies, and complements edit-distance metrics with an LLM-as-a-judge protocol. The central finding is that visual mis-recognition, not orthographic confusion, dominates errors (~56–60% for the strongest systems), with conjunct/hasanta errors contributing under 2%, challenging a long-standing assumption in Bengali OCR research.

Significance. The dataset construction is careful and well-documented: three independent native annotators, expert adjudication, strong inter-annotator agreement (Cohen's κ 0.81–0.93, mean pairwise CER 3.2%), bootstrapped confidence intervals, and transparent handling of the oracle P3 condition and of in-family judge bias. The dual verbatim/standard annotation is a valuable contribution that enables measuring silent over-correction and hallucination, which conventional OCR cannot exhibit. If the error-taxonomy claim were validated, the paper would provide an important, field-shaping result for Bengali OCR and VLM evaluation. However, the central claim currently rests on a string-only classifier whose 'visual' categories are not confirmed by any image-grounded evidence, so the headline result is not yet established. The benchmark and leaderboard are likely to be useful regardless.

major comments (3)
  1. [§6 and Appendix H (Table 2)] The error classifier receives only the (ground truth, prediction) string pair and explicitly has no access to the source image. The categories 'visual grapheme misread', 'lexical word substitution', and 'multi-character word corruption' are assigned from edit-distance thresholds and character-overlap rules, not from any perceptual evidence. Therefore the claim that ~56–60% of errors are 'visual mis-recognition' is not supported: the same string-edit pattern can arise from a vision-encoder miss, a language-prior replacement, a hallucination, or a decoding artifact. For instance, কপাট→েলাহ is counted as 'lexical word substitution' and described as a visual misreading, but the output shares no visual overlap with the source and is more plausibly a language-prior or hallucination error. The taxonomy needs validation against human cause-labeling on a sample of error pairs (with images), or th
  2. [§6, Table 2: visual:orthographic ratio] The headline visual-to-orthographic ratio (4.4× for Gemini Flash, 5.8× for Qwen3.5-9B) is computed from the unvalidated taxonomy. 'Lexical word substitution' is defined as any whole-word replacement with character overlap <40% or edit distance reaching 60% of word length; this captures any low-overlap error, regardless of cause. Because the classifier cannot distinguish visual errors from hallucination or language-prior errors, the ratio is not a meaningful estimate of the perceptual failure distribution. The subsequent conclusion that 'the bottleneck lies in the in-the-wild imagery rather than any specific model class' is a causal claim that the string-only classifier cannot support. A sensitivity analysis using stricter, visually grounded definitions, or a human-annotated cause sample, is required.
  3. [§6 and §7: conjunct/hasanta conclusion] The claim that the 'classical conjunct bottleneck has largely closed' (conjunct/hasanta errors under 2% for VLMs) is also dependent on the decision hierarchy in Appendix H. Any error that rewrites a word into a completely different form (e.g., a lost conjunct embedded in a hallucinated word) is routed to 'lexical word substitution' or 'semantic hallucination', never to 'conjunct/hasanta error'. The conjunct rate may therefore be artificially deflated by the classifier's priority order. At minimum, the paper should report how many errors with a hasanta/conjunct in the ground truth were reassigned to non-orthographic categories, or provide a per-glyph analysis of conjunct-containing ground-truth words.
minor comments (5)
  1. [Table 2] The taxonomy table reports only percentages of residual errors. Because the models have very different error rates (exact-match accuracy ranges from 0.16% to 35.6%), absolute counts or per-image rates would help assess the practical impact of each category.
  2. [Appendix H] The thresholds defining the categories (edit distance 2, character overlap <40%, edit distance ≥60% of length) are presented as fixed rules without justification or sensitivity analysis. A brief rationale or a test of neighboring thresholds would strengthen reproducibility.
  3. [§5] The claim of a 'persistent gap' in which larger models do not outperform smaller ones is based on only two Gemini models and three Qwen variants, which are also architecturally distinct. The paper appropriately cautions about this, but the wording in the abstract and conclusion ('persistent gap') is stronger than the evidence warrants.
  4. [§6] The qualitative examples used to support the visual-bottleneck claim (e.g., ফ্লয্াটিট→ফয্াক্টির, কপাট→েলাহ) are presented as self-evidently visual misreadings. Since the paper has the actual images, showing the source images (or a crowd-sourced human judgment of those specific cases) would be more convincing than relying on the reader's intuition.
  5. [Appendix H, word-boundary segmentation] The 'word-boundary segmentation' category includes whitespace-only normalization and a token split that leaves the character sequence intact. These events inflate WER but are not recognition failures; counting them separately from the visual/orthographic ratio would be cleaner.

Circularity Check

0 steps flagged

No significant circularity: benchmark measurements are operationalized a priori; the visual-dominance claim is a taxonomy summary, not a result derived from its own labels.

full rationale

This is a benchmark and evaluation paper, not a derivation chain. The central claim—that visual mis-recognition dominates Bengali scene-text errors—is an empirical summary of a deterministic, rule-based error classifier (Appendix H) whose categories were frozen before the full classification pass. No parameter is fitted to the target result, and the taxonomy's definitions are not derived from the measured error rates; rather, the rates are computed from the taxonomy. The LLM-as-a-judge protocol uses a Gemini model as primary judge, but the two Gemini systems are re-scored with Claude Sonnet 4.6 as an out-of-family judge, so the comparison does not reduce to in-family self-preference. The paper's most debatable point—that string-edit categories such as 'visual grapheme misread' correspond to perceptual causes without access to the source image—is a construct-validity limitation, not a circularity: the category definitions are independent of the observed outcome, and the paper explicitly discloses that the classifier receives only (ground truth, prediction) pairs. No step in the paper's argument is equivalent to its own inputs by construction, and no load-bearing premise rests on an unverified self-citation. Therefore the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The benchmark and taxonomy are measurement instruments, not derivations; the free parameters are the hand-chosen thresholds in the error classifier that shape the main empirical claim. The key domain assumptions are representativeness of the filtered data, the validity of string-edit categories as proxies for perceptual causes, and the judge's alignment with human judgment.

free parameters (4)
  • Lexical-substitution threshold = character overlap < 40% or edit distance >= 60% of word length
    Hand-picked threshold in Appendix H/Table 8; changing it changes the visual/orthographic error split that drives the headline finding.
  • Truncation length threshold = output length < 40% of ground-truth length
    Defines the truncation/no-output category in Stage 1 of the error classifier (Appendix H).
  • Semantic hallucination overlap threshold = character overlap < 25%
    Defines the semantic hallucination category in Stage 1; chosen by hand and affects the rate of this deployment-relevant category.
  • Homophone-class lookup table = fixed set such as ন/ণ, স/শ/ষ, ব/ভ
    Author-constructed lookup used to classify consonant homophone errors; determines the orthographic error share.
axioms (3)
  • domain assumption The benchmark retains only images that are readable, non-code-mixed, and not vulgar; this yields a representative sample of in-the-wild Bengali scene text.
    Section 3, Image Sourcing: 675 of 3,210 images were filtered. Representativeness is assumed, not measured.
  • domain assumption The string-edit error taxonomy (Appendix H) correctly maps edit patterns to perceptual causes (visual vs orthographic).
    The classifier has no access to the source image, so the category 'visual mis-recognition' is inferred from edit statistics, not from image inspection.
  • domain assumption The LLM-as-a-judge (Gemini 2.5 Pro, temperature 0) provides a meaning-aware score that tracks human judgment.
    Validated on 200 pairs with Spearman rho = 0.89, but the judge is itself a frontier VLM and may share blind spots (Appendix F).

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models." pith.science (2026). https://pith.science/paper/GCBLWQKO

@misc{pith2026260803884,
  author       = {Pith},
  title        = {Pith review of: BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GCBLWQKO}},
  note         = {Machine review of arXiv:2608.03884}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

In-the-wild Bengali scene text recognition is largely unmeasured: existing resources target handwritten documents or constrained sign-board parsing, report only aggregate edit-distance metrics, and evaluate either conventional OCR or VLMs, never both on the same in-the-wild data. To address this gap, we introduce BANGLAWILD, a benchmark of 2,535 Bengali scene text images, each paired with a verbatim gold transcription, two categorical axes, four diagnostic attributes, and an orthographically standard form where the in-image text deviates from canonical spelling. We evaluate fifteen VLMs and three conventional OCR systems under three prompting strategies, fine-tune 6 open-source models with LoRA, and complement edit-distance metrics with an LLM-as-a-Judge evaluation. Our results reveal a persistent gap in which larger models within the same family do not outperform smaller ones. Our fifteen-class error taxonomy shows that visual mis-recognition accounts for ~60% of errors in the strongest systems, while conjunct-related errors contribute under 2%, challenging a long-standing assumption in Bengali OCR research; the same visual dominant profile also holds across architectures, including the one conventional baseline that reads Bengali reliably. Prompt language mainly affects cross-script drift and LoRA reduces catastrophic failures in weak models without lifting the ceiling on already competent ones. Code and data will be publicly released.

Figures

Figures reproduced from arXiv: 2608.03884 by Eshat Tanzeem, Mir Md Inzamam, Sadab Shiper, Tawsif Tashwar Dipto.

Figure 1
Figure 1. Figure 1: Category-wise distribution of BANGLAWILD across CONTEXT (outer ring) and CONTENT (inner ring) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the BANGLAWILD Data Collection and Annotation Pipeline. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

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

90 extracted references · 53 canonical work pages · 8 internal anchors

  1. [1]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  2. [2]

    Publications Manual , year = "1983", publisher =

    1983

  3. [3]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    arXiv 1981

  4. [4]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  5. [5]

    Dan Gusfield , title =. 1997

    1997

  6. [6]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  7. [7]

    , journal =

    Gunna, Sanjana and Saluja, Rohit and Jawahar, C.\ V. , journal =. Transfer Learning for Scene Text Recognition in

  8. [8]

    arXiv preprint arXiv:2502.06172 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Maeda, Koki and Okazaki, Naoaki , journal =

  10. [10]

    Seeing the Signs: A Survey of Edge-Deployable OCR Models for Billboard Visibility Analysis

    Seeing the Signs: A Survey of Edge-Deployable. arXiv preprint arXiv:2507.11730 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual VLMs for Bengali Culture Understanding Across Historically Linked Languages and Regional Dialects

    Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual. arXiv preprint arXiv:2603.21165 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    arXiv preprint arXiv:2603.15949 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    Sohel and Shahriyar, Rifat , booktitle =

    Bhattacharjee, Abhik and Hasan, Tahmid and Ahmad, Wasi and Mubasshir, Kazi Samin and Islam, Md Saiful and Iqbal, Anindya and Rahman, M. Sohel and Shahriyar, Rifat , booktitle =. 2022 , pages =

    2022

  14. [14]

    arXiv preprint arXiv:2404.18930 , year =

    Hallucination of Multimodal Large Language Models: A Survey , author =. arXiv preprint arXiv:2404.18930 , year =

    Pith/arXiv arXiv

  15. [15]

    arXiv preprint arXiv:2507.19024 , year =

    A Survey of Multimodal Hallucination Evaluation and Detection , author =. arXiv preprint arXiv:2507.19024 , year =

    arXiv

  16. [16]

    arXiv preprint arXiv:2602.22918 , year =

    Where Vision Becomes Text: Locating the. arXiv preprint arXiv:2602.22918 , year =

    Pith/arXiv arXiv

  17. [17]

    Advances in Neural Information Processing Systems , volume =

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author =. Advances in Neural Information Processing Systems , volume =

  18. [18]

    CVPR , year =

    Compositional Chain-of-Thought Prompting for Large Multimodal Models , author =. CVPR , year =

  19. [19]

    arXiv preprint arXiv:2405.13872 , year =

    Image-of-Thought Prompting for Visual Reasoning Refinement in Multimodal Large Language Models , author =. arXiv preprint arXiv:2405.13872 , year =

    Pith/arXiv arXiv

  20. [20]

    Advances in Neural Information Processing Systems , volume =

    Flamingo: a Visual Language Model for Few-Shot Learning , author =. Advances in Neural Information Processing Systems , volume =

    work page

  21. [21]

    Advances in Neural Information Processing Systems , volume =

    Multimodal Few-Shot Learning with Frozen Language Models , author =. Advances in Neural Information Processing Systems , volume =

    work page

  22. [22]

    Quality and Quantity , volume =

    Measuring the Reliability of Qualitative Text Analysis Data , author =. Quality and Quantity , volume =

    work page

  23. [23]

    Australian & New Zealand Journal of Statistics , volume =

    Permutation Tests for Linear Models , author =. Australian & New Zealand Journal of Statistics , volume =

    work page

  24. [24]

    Communications of the ACM , volume =

    Datasheets for Datasets , author =. Communications of the ACM , volume =

  25. [25]

    Transactions of the Association for Computational Linguistics , volume =

    Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science , author =. Transactions of the Association for Computational Linguistics , volume =

    work page

  26. [26]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

  27. [27]

    arXiv preprint arXiv:2511.21631 , year =

    Pith/arXiv arXiv

  28. [28]

    Zhu, Jinguo and Wang, Weiyun and Chen, Zhe and Liu, Zhaoyang and Ye, Shenglong and Gu, Lixin and Tian, Hao and Duan, Yuchen and Su, Weijie and Shao, Jie and others , journal =

    work page

  29. [29]

    2025 , howpublished =

    The. 2025 , howpublished =

    2025

  30. [30]

    Towards Detecting, Recognizing, and Parsing the Address Information from

    Murad, Hasan and Ali, Mohammed Eunus , journal =. Towards Detecting, Recognizing, and Parsing the Address Information from

    work page

  31. [31]

    Advances and Challenges in Multilingual

    [Author Last] and [Author Last] , journal =. Advances and Challenges in Multilingual. 2023 , volume =

    work page 2023

  32. [32]

    [Authors] , journal =

    work page

  33. [33]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    work page

  34. [34]

    Seeing the Signs: A Survey of Edge-Deployable

    [Authors] , journal =. Seeing the Signs: A Survey of Edge-Deployable

    work page

  35. [35]

    arXiv preprint arXiv:2510.12287 , year =

    Vision Language Models Map Logos to Text via Semantic Entanglement in the Visual Projector , author =. arXiv preprint arXiv:2510.12287 , year =

    work page arXiv

  36. [36]

    Sprague, Zayne and others , journal =. To

    work page

  37. [37]

    2017 , url =

    Bengali Romanization Table , author =. 2017 , url =

    work page 2017

  38. [38]

    arXiv preprint arXiv:2403.12736 , year =

    Towards Multimodal In-Context Learning for Vision & Language Models , author =. arXiv preprint arXiv:2403.12736 , year =

    Pith/arXiv arXiv

  39. [39]

    2024 , howpublished =

    Bengali Layout Requirements , author =. 2024 , howpublished =

    work page 2024

  40. [40]

    2024 , howpublished =

    Graffiti vs Regime , author =. 2024 , howpublished =

    work page 2024

  41. [41]

    2024 , howpublished =

    What is happening at the quota-reform protests in. 2024 , howpublished =

    work page 2024

  42. [42]

    Computers and Electrical Engineering , volume =

    A Crowdsource Based Framework for Bengali Scene Text Data Collection and Detection , author =. Computers and Electrical Engineering , volume =. 2023 , publisher =

    work page 2023

  43. [43]

    Bharat Scene Text: A Novel Comprehensive Dataset and Benchmark for Indian Language Scene Text Understanding

    Bharat Scene Text: A Novel Comprehensive Dataset and Benchmark for Indian Language Scene Text Understanding , author=. arXiv preprint arXiv:2511.23071 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    Tang, Jingqun and Liu, Qi and Ye, Yongjie and Lu, Jinghui and Wei, Shu and Lin, Chunhui and Li, Wanqing and others , journal =

    work page

  45. [45]

    Nonesung, Surapon and others , booktitle =

    work page

  46. [46]

    TituLLMs Team , journal =

    work page

  47. [47]

    Chen, Zhe and Wu, Jiannan and Wang, Wenhai and Su, Weijie and Chen, Guo and Xing, Sen and Zhong, Muyan and Zhang, Qinglong and Zhu, Xizhou and Lu, Lewei and others , booktitle =

    work page

  48. [48]

    Zhu, Jinguo and Wang, Weiyun and Chen, Zhe and Liu, Zhaoyang and Ye, Shenglong and others , journal =

    work page

  49. [49]

    Wang, Weiyun and Gao, Zhangwei and Gu, Lixin and Pu, Hengjun and Cui, Long and Wei, Xingguang and others , journal =

    work page

  50. [50]

    ArXiv , year=

    Gemma 3 Technical Report , author=. ArXiv , year=

  51. [51]

    Bai, Jinze and Bai, Shuai and Yang, Shusheng and Wang, Shijie and Tan, Sinan and Wang, Peng and Lin, Junyang and Zhou, Chang and Zhou, Jingren , journal =

    work page

  52. [52]

    Wang, Peng and Bai, Shuai and Tan, Sinan and Wang, Shijie and Fan, Zhihao and Bai, Jinze and Chen, Keqin and others , journal =

    work page

  53. [53]

    and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =

    Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =

  54. [54]

    Dettmers, Tim and Pagnoni, Artidoro and Holtzman, Ari and Zettlemoyer, Luke , booktitle =

  55. [55]

    and Khaneja, Ayush and Wagholikar, Kavishwar B

    Shakya, Prabin R. and Khaneja, Ayush and Wagholikar, Kavishwar B. , journal =. For clinical data extraction,

    work page

  56. [56]

    , journal =

    Zhu, et al. , journal =. Towards Minimal Fine-Tuning of

    work page

  57. [57]

    , journal =

    Pipatanakul, et al. , journal =. Typhoon

    work page

  58. [58]

    arXiv preprint arXiv:2503.19786 , year =

    Gemma 3 Technical Report , author =. arXiv preprint arXiv:2503.19786 , year =

    Pith/arXiv arXiv

  59. [59]

    arXiv preprint arXiv:2505.09388 , year =

    Qwen3 Technical Report , author =. arXiv preprint arXiv:2505.09388 , year =

    Pith/arXiv arXiv

  60. [60]

    arXiv preprint arXiv:2604.15804 , year =

    Qwen3.5-Omni Technical Report , author =. arXiv preprint arXiv:2604.15804 , year =

    Pith/arXiv arXiv

  61. [61]

    2025 , url =

    Wang, Weiyun and Gao, Zhangwei and Gu, Lixin and Pu, Hengjun and Cui, Long and Wei, Xingguang and Liu, Zhaoyang and Jing, Linglin and Ye, Shenglong and Shao, Jie and others , journal =. 2025 , url =

    work page 2025

  62. [62]

    2025 , howpublished =

    Claude Haiku 4.5 System Card , author =. 2025 , howpublished =

    work page 2025

  63. [63]

    2026 , howpublished =

    Claude Sonnet 4.6 System Card , author =. 2026 , howpublished =

    work page 2026

  64. [64]

    arXiv preprint arXiv:2507.06261 , year =

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities , author =. arXiv preprint arXiv:2507.06261 , year =

    Pith/arXiv arXiv

  65. [65]

    2025 , howpublished =

    Gemini 3 Flash is now available in Gemini CLI , author =. 2025 , howpublished =

    work page 2025

  66. [66]

    arXiv preprint arXiv:2601.03267 , year =

    Pith/arXiv arXiv

  67. [67]

    2026 , howpublished =

    Introducing. 2026 , howpublished =

    2026

  68. [68]

    arXiv preprint arXiv:2601.11659 , year =

    The. arXiv preprint arXiv:2601.11659 , year =

    arXiv

  69. [69]

    NeurIPS Datasets & Benchmarks , year =

    David Romero and others , title =. NeurIPS Datasets & Benchmarks , year =

    work page

  70. [70]

    arXiv preprint arXiv:2410.21200 , year =

    Abdullah Khan Zehady and Safi Al Mamun and Naymul Islam and Santu Karmaker , title =. arXiv preprint arXiv:2410.21200 , year =

    work page arXiv

  71. [71]

    Proc.\ ACL (Short Papers) , year =

    Md Nishat Raihan and Marcos Zampieri , title =. Proc.\ ACL (Short Papers) , year =

    work page

  72. [72]

    Bangla-Bayanno: A 52K-Pair Bengali Visual Question Answering Dataset with LLM-Assisted Translation Refinement

    Md.\ Hasan and others , title =. arXiv preprint arXiv:2508.19887 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  73. [73]

    ChitroJera: A Regionally Relevant Visual Question Answering Dataset for Bangla

    Deeparghya Dutta Barua and Pranabesh Sourove and Yusuf Fahim and others , title =. arXiv preprint arXiv:2410.14991 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  74. [74]

    2019 , pages =

    Nayef, Nibal and Patel, Yash and Busta, Michal and Chowdhury, Pinaki Nath and Karatzas, Dimosthenis and Khlif, Wafa and Matas, Jiri and Pal, Umapada and Burie, Jean-Christophe and Liu, Cheng-lin and Ogier, Jean-Marc , booktitle =. 2019 , pages =

    work page 2019

  75. [75]

    EMNLP , year =

    Yifan Li and Yifan Du and Kun Zhou and Jinpeng Wang and Wayne Xin Zhao and Ji-Rong Wen , title =. EMNLP , year =

    work page

  76. [76]

    Transactions on Machine Learning Research , year =

    Zhuosheng Zhang and Aston Zhang and Mu Li and Hai Zhao and George Karypis and Alex Smola , title =. Transactions on Machine Learning Research , year =

  77. [77]

    arXiv preprint arXiv:2503.12605 , year =

    Yaoting Wang and Shengqiong Wu and Yuecheng Zhang and others , title =. arXiv preprint arXiv:2503.12605 , year =

    Pith/arXiv arXiv

  78. [78]

    2025 , howpublished =

    Mohammed, Anwer and Hasan, Md Hasib , title =. 2025 , howpublished =. doi:10.17632/fxw9d2g5z4.1 , url =

    work page doi:10.17632/fxw9d2g5z4.1 2025

  79. [79]

    OCRBench: on the hidden mystery of OCR in large multimodal models , volume=

    Liu, Yuliang and Li, Zhang and Huang, Mingxin and Yang, Biao and Yu, Wenwen and Li, Chunyuan and Yin, Xu-Cheng and Liu, Cheng-Lin and Jin, Lianwen and Bai, Xiang , year=. OCRBench: on the hidden mystery of OCR in large multimodal models , volume=. Science China Information Sciences , publisher=. doi:10.1007/s11432-024-4235-6 , number=

    doi:10.1007/s11432-024-4235-6

  80. [80]

    von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin , license =

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.