Pith. sign in

REVIEW 5 major objections 5 minor 50 references

TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that injecting ground-truth label embeddings into a scene text recognizer's decoder during training, then masking them with a loss-aware rule, improves accuracy on six benchmarks with no inference-time cost.

desk verdict A plausible training-time curriculum idea for STR whose reported evidence is undercut by inconsistent tables and benchmark-tuned hyperparameters; the method may still work, but the paper needs major revision. read the letter →

arxiv 2508.01153 v1 pith:EBC2K224 submitted 2025-08-02 cs.CV

Xiahan Yang , Hui Zheng This is my paper
classification cs.CV
keywords SceneTextRecognitionCurriculumLearningTraining-timelabelinjectionLoss-awaremaskingEncoder-decodermodelsBenchmarkevaluationModel-agnostictrainingVision-onlyinference
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

TEACH is a training-only scheme for scene text recognition that gives the decoder the correct word during training and then gradually takes it away. The authors claim that starting from full label guidance and masking the labels according to the recent training loss forces the model to learn stronger visual features, so that at test time, when no label is available, it recognizes text more accurately. Across six standard benchmarks, models equipped with TEACH improve on their base counterparts under both synthetic and real training data, with the largest gains on distorted, occluded, and low-quality text. The method is positioned as a lightweight alternative to language-model-based recognizers because it needs no external pretraining and adds no inference-time computation.

What carries the argument

The central object is the loss-aware masking rule $r = \max(0, \min(1, \alpha\cdot(\mathrm{Loss}-\beta)))$, which sets the fraction of label tokens replaced by a padding token in each training step. High loss means little masking, so label guidance stays on; low loss means heavy masking, and once loss falls below $\beta$ the label stream is fully padded out. This rule is what turns a fixed label-injection trick into a curriculum, letting the decoder shift from reconstructing the label embedding to reading visual features, while keeping the ability to relax masking if loss spikes on hard samples. The companion mechanism is the concatenation of text embeddings with visual tokens in the decoder input, so the model must decide which stream to trust at each step.

What would settle it

Fix $\alpha$ and $\beta$ on a validation split drawn only from training data, then evaluate on the six benchmarks; if the TEACH models no longer beat their baselines by a margin beyond repeated-run noise, the out-of-sample claim fails. A second check: train with randomly permuted ground-truth labels under the same schedule; if accuracy stays as high as with correct labels, the effect is not semantic guidance.

Watch

Extended reading notes

Core claim

On its own terms, the central claim is that a curriculum of text hints can bootstrap visual recognition: during training the ground-truth label sequence is embedded and concatenated with the visual tokens in the decoder input, and this label stream is masked according to $r = \max(0, \min(1, \alpha\cdot(\mathrm{Loss}-\beta)))$, where Loss is the previous batch's loss. High loss keeps the labels visible; low loss hides them; once loss falls below $\beta$ the labels vanish entirely. The authors report that this schedule consistently improves word accuracy over the ViTSTR and PARSeq baselines on IIIT5K, SVT, IC13, IC15, SVTP, and CUTE, with the largest improvements on blurry, distorted, and occluded samples, and that PARSeq+TEACH reaches state-of-the-art results on five of six synthetic-trained benchmarks and on IC15, SVTP, and CUTE with real-data training. They further report gains on ArT, COCO-Text, and Uber-Text and on the recurrent CRNN architecture, supporting the claim that the strategy is model-agnostic; at inference the model takes only the image, so no extra computation or parameters are added.

Load-bearing premise

The load-bearing premise is that the benchmark numbers honestly measure what the method would do on new data, since the two masking hyperparameters were tuned with the evaluation benchmarks in view (the grid-search table and a baseline's stabilized loss), and if that tuning leaked into the headline results, the reported generalization gains could shrink on unseen test sets.

Editorial extensions

If this is right

  • TEACH can be plugged into existing encoder-decoder STR models without changing their networks, adding parameters, or altering inference, so any future backbone can inherit the recipe directly.
  • Models trained with synthetic data get a consistent lift: the paper reports ViTSTR+TEACH gaining up to 1.7 points on IC13 and 1.6 points on CUTE, and PARSeq+TEACH reaching 97.5% on IIIT5K.
  • The same pattern holds under real-data training, where PARSeq+TEACH reaches 99.2% on IIIT5K and 97.6% on CUTE and surpasses strong baselines such as ABINet and MAERec on several benchmarks.
  • Because loss-aware masking beats linear decay in the ablation, the adaptive schedule, not mere label injection, is the load-bearing component.

Reading between the lines

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

  • If the evaluation honestly measures out-of-sample behavior, the method implies that other structured prediction tasks, such as handwritten text recognition, image captioning, or speech recognition, could use a similar loss-aware ground-truth hint curriculum that vanishes at test time.
  • The reported hyperparameter search is a caution flag: $\alpha$ and $\beta$ were chosen over the same six benchmark sets, with $\beta$ tied to a baseline's stabilized loss, so the true out-of-sample gain is likely smaller than the headline tables suggest until the schedule is fixed on a held-out split.
  • A testable extension of the paper's own logic is to train with permuted labels under the same masking rule; if accuracy stays as high as with correct labels, the benefit comes from regularization or noise tolerance rather than semantic alignment.
  • Inspecting decoder attention on padding positions during late training could show whether the model truly reads vision-only or merely learns a constant bias from the padded label slots.
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

5 major / 5 minor

Summary. The paper proposes TEACH, a training-only curriculum method for scene text recognition that injects ground-truth label embeddings into the decoder input and progressively masks them using a loss-dependent schedule. The authors claim consistent accuracy improvements over ViTSTR and PARSeq baselines without inference overhead, across synthetic and real training data and several benchmarks. The core idea is simple and architecture-agnostic, but the empirical evidence as presented contains internal contradictions (Table 1 SVT drop, duplicate ViTSTR-S rows in Table 2, reused caption in Table 3), an under-specified masking mechanism (Eqs. 4-5), and hyperparameter selection directly on the evaluation benchmarks (Table 6).

Significance. The method has practical appeal: it requires no external language models, no inference-time changes, and could be dropped into existing encoder-decoder STR models. The paper evaluates on a broad set of benchmarks and two base architectures, and the qualitative examples suggest the mechanism can help on distorted or low-contrast text. However, the current manuscript does not establish the central empirical claim. The test-set tuning of alpha and beta means the reported gains may not generalize, and the table inconsistencies prevent verification even in-sample. With corrected experiments and clean reporting, TEACH could be a useful contribution as a lightweight training augmentation; as submitted, its significance is not yet demonstrated.

major comments (5)
  1. [Section 3.2, Eq. (5)] The definition and interpretation of the masking rate r are internally inconsistent. The text states that 'A high loss results in low masking (more label input), while a low loss increases masking', but Eq. (5) defines r as increasing in Loss. Moreover, the final stage says that when the loss falls below beta the label input is fully removed (r = 0 under Eq. (5)), which is the opposite of what a 'masking rate' of 0 should mean. In addition, Eq. (4) uses a binary mask M in {0,1}^{Lt} but r is a scalar, and the paper never specifies how the scalar r produces M. Because the loss-aware curriculum is the central mechanism, this ambiguity blocks implementation and evaluation of the method.
  2. [Table 2] The real-data comparison table contains two identical 'ViTSTR-S' baseline rows with different numbers (97.9/96/97.8/89/87.5/91.5/96.2 vs 98.0/95.0/97.2/88.3/87.4/91.8/97.6) and no distinguishing label, and the header includes two 'IC15' columns without indicating which variant (1811 vs 2077) each refers to. Under these rows, ViTSTR+TEACH is below at least one baseline on IIIT5K, IC13, both IC15 columns, and CUTE. The claim in Section 4.3 that 'ViTSTR+TEACH improves on IC15 and SVTP by 0.6% and 0.8%' cannot be reproduced from either baseline row. This makes the central claim of consistent improvement unverifiable for the real-data setting.
  3. [Table 1] The synthetic-data results contradict the headline claim of consistent improvement: ViTSTR+TEACH scores 91.3 on SVT against 93.8 for ViTSTR-S, a drop of 2.5 percentage points, while the caption and Section 4.3 state that TEACH 'improves both ViTSTR and PARSeq baselines across all benchmarks.' This is not a minor fluctuation; it directly refutes the word 'consistent' in the central claim.
  4. [Section 4.4, Table 6] The hyperparameters alpha and beta are selected by grid search over accuracy averaged across the six evaluation benchmarks (Table 6), and the selected configuration (alpha=2, beta=0.1) is used for the headline numbers in Tables 1-3. Since the same benchmarks are used for both selection and final reporting, the reported improvements are not out-of-sample estimates. Additionally, Section 4.2 states that beta is set 'slightly below the stabilized loss' of a pretrained baseline, which is a different criterion from the grid search; the manuscript does not reconcile these procedures. The claimed general improvement therefore needs to be re-established with a held-out validation set or with a clearly specified selection protocol that does not use the test benchmarks.
  5. [Table 3] Table 3's caption is copied from Table 2, mentioning 'six STR benchmarks', 'DiG-ViT, ABINet, and MAERec', and 'five out of six benchmarks', but the table actually reports four columns (ArT, COCO-Text, Uber-Text, Total) and contains none of those models. Since this table is the evidence for the 'challenging large-scale benchmarks' claim, the mismatch prevents the reader from knowing what was evaluated and how the numbers were computed.
minor comments (5)
  1. [Section 4.1] The definitions of 'ALL benchmark set' and 'Clean subset' are not used consistently with the tables; please label the duplicate IC13/IC15 columns explicitly (e.g., IC15-1811 and IC15-2077) in every table.
  2. [Section 3.4 and Table 7] Section 3.4 says CRNN is excluded from the main analysis because 'preliminary experiments found it difficult to propagate useful gradients', yet Table 7 reports CRNN+TEACH gains; please clarify why CRNN results are not in the main tables or remove the preliminary-experiment statement.
  3. [Section 3.2] Please provide the exact mapping from the scalar masking rate r in Eq. (5) to the binary mask M in Eq. (4), including how the sequence-length dimension is handled (e.g., Bernoulli sampling, token-level thresholding, or per-position masking).
  4. [Figure 3] The word 'Predication' appears in the figure text; it should be 'Prediction'.
  5. [Section 4.3] In the qualitative comparison, the notation alternates between 'ViTSTR-TEACH' and 'ViTSTR+TEACH' and between 'PARSeq-TEACH' and 'PARSeq+TEACH'; please standardize and clarify which variant (e.g., PARSeqA vs PARSeqN) is used.

Circularity Check

1 steps flagged · score 5.0 of 10

TEACH's training scheme is not circular, but the headline benchmark gains are partly selected by tuning alpha/beta on the same six test benchmarks, so the generalization claim is not fully out-of-sample.

  1. fitted input called prediction [Section 4.4 (Ablation Analysis), Table 6; Section 4.2 (Implementation Details)]
    "To investigate the robustness of TEACH under different schedules, we conduct an ablation study on combinations of α ∈ {0.5, 1, 2} and β ∈ {0.01, 0.05, 0.1} using the PARSeq backbone. Table 6 summarizes results on both synthetic and real training data. We observe that TEACH consistently improves over the baseline across most settings. In particular, a larger α combined with a higher β (i.e., slower masking decay) tends to yield better results, with the best performance achieved when α/β = 2/0.1."

    The masking schedule's hyperparameters (α, β) are selected by grid search to maximize accuracy averaged over the same six benchmarks (IIIT5K, SVT, IC13, IC15, SVTP, CUTE) that Tables 1-3 use as evidence of TEACH's 'consistently improved accuracy.' Section 4.2 also sets β from the pretrained baseline's stabilized loss. Choosing the schedule on the evaluation set means the reported gains are not independent out-of-sample predictions of generalization; the headline numbers are partly a product of maximizing the exact metric used to claim success. This is not a full reduction by construction because the model must still learn visual features after labels are removed, but it makes the generalization claim statistically forced on the tuned benchmarks.

full rationale

The core derivation chain is self-contained: ground-truth embeddings are injected at training time, masked by a loss-dependent rule, and removed at inference, so the model is forced to rely on vision at test time. No load-bearing step is justified only by the authors' prior work, and the training objective is not equivalent to the reported accuracy. The one genuine circularity-adjacent defect is the hyperparameter selection in Section 4.4: α and β are chosen on the same six evaluation benchmarks later presented as validation, so the 'consistent improvements' in Tables 1-3 are partly in-sample to the hyperparameter choice. Separately, the paper has verifiability problems that are not circularity but affect trust in the printed deltas: Table 2 lists the ViTSTR-S Real(3.3M) baseline twice with different accuracies, the ViTSTR+TEACH row is not consistent with the claimed +0.6%/+0.8% gains, and Table 3 reuses Table 2's caption. These are data-integrity issues rather than derivation-by-construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

TEACH introduces no new physical or architectural entity; it is a training schedule over existing embeddings. The label embedding and padding vector are standard learned tensors, not new theoretical objects.

free parameters (3)
  • alpha (masking sensitivity) = 2 (best in Table 6 ablation)
    Scales loss into masking rate in Eq. 5. Chosen by grid search over {0.5, 1, 2} using average accuracy on the six evaluation benchmarks, so it is fitted to the reported test sets.
  • beta (loss threshold) = 0.1 (best in Table 6 ablation)
    Threshold in Eq. 5 below which labels are fully masked. Section 4.2 also suggests setting it just below the stabilized baseline loss; the value is selected on benchmark accuracy rather than on a held-out set.
  • padid (padding embedding) = learned, value not reported
    Fixed or learnable placeholder used in Eq. 4 to replace masked label tokens. Its trained value is not analyzed, but it is part of the injected input distribution.
assumptions (4)
  • ad hoc to paper The loss-aware masking schedule with hand-set alpha and beta provides a beneficial curriculum for learning visual features.
    Introduced in Section 3.2, Eq. 5; no theory or independent validation supports this, and the schedule is tuned on the evaluation benchmarks.
  • domain assumption Following the official training configurations of PARSeq and ViTSTR with the same epoch count yields fair baselines.
    Section 4.2; no code, logs, or checkpoints are provided, and Table 2 lists two different ViTSTR-S rows with different numbers, so the baselines are not verifiable.
  • domain assumption The six evaluation benchmarks are held out during development and used only for final assessment.
    Standard protocol, but violated in Section 4.4 where alpha and beta are chosen by accuracy on these same benchmarks.
  • ad hoc to paper The scalar masking rate r in Eq. 5 can be turned into the binary mask M in Eq. 4 without changing model behavior.
    The paper never describes how r maps to M; this is an implicit and underspecified modeling decision.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition." pith.science (2026). https://pith.science/paper/EBC2K224

@misc{pith2026250801153,
  author       = {Pith},
  title        = {Pith review of: TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EBC2K224}},
  note         = {Machine review of arXiv:2508.01153}
}
read the original abstract

Scene Text Recognition (STR) remains a challenging task due to complex visual appearances and limited semantic priors. We propose TEACH, a novel training paradigm that injects ground-truth text into the model as auxiliary input and progressively reduces its influence during training. By encoding target labels into the embedding space and applying loss-aware masking, TEACH simulates a curriculum learning process that guides the model from label-dependent learning to fully visual recognition. Unlike language model-based approaches, TEACH requires no external pretraining and introduces no inference overhead. It is model-agnostic and can be seamlessly integrated into existing encoder-decoder frameworks. Extensive experiments across multiple public benchmarks show that models trained with TEACH achieve consistently improved accuracy, especially under challenging conditions, validating its robustness and general applicability.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 47 canonical work pages

  1. [1]

    TEACH: Text Encoding as Curriculum Hints for Scene Text Recognition

    INTRODUCTION Scene Text Recognition (STR), a key task in Optical Char- acter Recognition (OCR), aims to transcribe text from nat- ural scene images. Compared to document OCR, STR in- volves a broader range of challenges, including perspective distortions, background clutter, uneven lighting, occlusions, and variations in fonts, styles, and languages. Thes...

    work page Pith review arXiv 2025

  2. [2]

    TYPO” as “TYPE

    RELA TED WORK Scene Text Recognition (STR) methods can be broadly grouped into three major paradigms: convolutional-recurrent architectures, transformer-based models, and vision-language multimodal approaches. CRNN and Recurrent Models. Convolutional Recurrent Neural Networks (CRNN) [1] are early yet influential archi- tectures in STR. They use CNNs to ex...

    work page

  3. [3]

    This simulates a curriculum-style learn- ing process where the model first relies on explicit textual supervision, then gradually learns to infer based purely on visual cues

    METHODS We propose TEACH, a plug-and-play training strategy for scene text recognition (STR) that injects ground-truth label embeddings into the model input and progressively removes them during training. This simulates a curriculum-style learn- ing process where the model first relies on explicit textual supervision, then gradually learns to infer based ...

    work page

  4. [4]

    Experiments Setup Following standard practice in scene text recognition (STR) [19, 5], we adopt both synthetic and real-world datasets for train- ing and evaluation

    EXPERIMENTS 4.1. Experiments Setup Following standard practice in scene text recognition (STR) [19, 5], we adopt both synthetic and real-world datasets for train- ing and evaluation. Synthetic datasets. We use two widely adopted synthetic datasets to pretrain STR models: MJSynth (MJ) [27], con- taining approximately 9 million rendered word images, and Syn...

    work page 2013

  5. [5]

    TEACH injects ground-truth labels into the input sequence during training and gradually masks them based on model performance

    CONCLUSION In this work, we proposed TEACH, a lightweight and model- agnostic training strategy for scene text recognition (STR). TEACH injects ground-truth labels into the input sequence during training and gradually masks them based on model performance. This progressive supervision simulates a cur- riculum learning process, enabling the model to shift ...

    work page

  6. [6]

    An end-to- end trainable neural network for image-based sequence recognition and its application to scene text recogni- tion,

    Baoguang Shi, Xiang Bai, and Cong Yao, “An end-to- end trainable neural network for image-based sequence recognition and its application to scene text recogni- tion,” IEEE transactions on pattern analysis and ma- chine intelligence, vol. 39, no. 11, pp. 2298–2304, 2016

    work page 2016

  7. [7]

    Attention is all you need,

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin, “Attention is all you need,” Ad- vances in neural information processing systems , vol. 30, 2017

    work page 2017

  8. [8]

    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, Sylvain Gelly, et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020

    arXiv 2010

Show all 50 references
  1. [9]

    Vision transformer for fast and effi- cient scene text recognition,

    Rowel Atienza, “Vision transformer for fast and effi- cient scene text recognition,” in International Confer- ence on Document Analysis and Recognition . Springer, 2021, pp. 319–334

    2021

  2. [10]

    Scene text recog- nition with permuted autoregressive sequence models,

    Darwin Bautista and Rowel Atienza, “Scene text recog- nition with permuted autoregressive sequence models,” in European Conference on Computer Vision, Cham, 10 2022, pp. 178–196, Springer Nature Switzerland

    2022

  3. [11]

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

    Minghao Li, Tengchao Lv, Jingye Chen, Lei Cui, Yi- juan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, and Furu Wei, “Trocr: Transformer-based optical character recognition with pre-trained models,” in Proceedings of the AAAI Conference on Artificial Intelligence, 2023, vol. 37, ...

    2023

  4. [12]

    Read like humans: Au- tonomous, bidirectional and iterative language model- ing for scene text recognition,

    Shancheng Fang, Hongtao Xie, Yuxin Wang, Zhendong Mao, and Yongdong Zhang, “Read like humans: Au- tonomous, bidirectional and iterative language model- ing for scene text recognition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, ...

    2021

  5. [13]

    Dtrocr: Decoder-only transformer for optical character recognition,

    Masato Fujitake, “Dtrocr: Decoder-only transformer for optical character recognition,” in Proceedings of the IEEE/CVF Winter Conference on Applications of Com- puter Vision, 2024, pp. 8025–8035

    2024

  6. [14]

    Clip4str: A simple baseline for scene text recognition with pre-trained vision-language model,

    Shuai Zhao, Xiaohan Wang, Linchao Zhu, and Yi Yang, “Clip4str: A simple baseline for scene text recognition with pre-trained vision-language model,” arXiv preprint arXiv:2305.14014, 2023

    2023 arXiv

  7. [15]

    Curriculum learning,

    Yoshua Bengio, J ´erˆome Louradour, Ronan Collobert, and Jason Weston, “Curriculum learning,” in ICML, 2009

    2009

  8. [16]

    Self-paced learning for latent variable models,

    M. Pawan Kumar, Benjamin Packer, and Daphne Koller, “Self-paced learning for latent variable models,” in NeurIPS, 2010

    2010

  9. [17]

    Recursive recurrent nets with attention modeling for ocr in the wild,

    Chen-Yu Lee and Simon Osindero, “Recursive recurrent nets with attention modeling for ocr in the wild,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2231–2239

    2016

  10. [18]

    Gcrnn: graph convolutional recurrent neural network for compound–protein interaction prediction,

    Ermal Elbasani, Soualihou Ngnamsie Njimbouom, Tae- Jin Oh, Eung-Hee Kim, Hyun Lee, and Jeong-Dong Kim, “Gcrnn: graph convolutional recurrent neural network for compound–protein interaction prediction,” BMC bioinformatics, vol. 22, no. 5, pp. 1–14, 2021

    2021

  11. [19]

    Ma-crnn: a multi-scale attention crnn for chinese text line recognition in natural scenes,

    Guofeng Tong, Yong Li, Huashuai Gao, Huairong Chen, Hao Wang, and Xiang Yang, “Ma-crnn: a multi-scale attention crnn for chinese text line recognition in natural scenes,” International Journal on Document Analysis and Recognition (IJDAR), vol. 23, pp. 103–114, 2020

    2020

  12. [20]

    Aster: An attentional scene text recognizer with flexible rectifica- tion,

    Baoguang Shi, Mingkun Yang, Xinggang Wang, Pengyuan Lyu, Cong Yao, and Xiang Bai, “Aster: An attentional scene text recognizer with flexible rectifica- tion,” IEEE transactions on pattern analysis and ma- chine intelligence, vol. 41, no. 9, pp. 2035–2048, 2018

    2018

  13. [21]

    Towards accu- rate scene text recognition with semantic reasoning net- works,

    Deli Yu, Xuan Li, Chengquan Zhang, Tao Liu, Junyu Han, Jingtuo Liu, and Errui Ding, “Towards accu- rate scene text recognition with semantic reasoning net- works,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2020, pp. 12113–12122

    2020

  14. [22]

    Textscanner: Reading characters in or- der for robust scene text recognition,

    Zhaoyi Wan, Minghang He, Haoran Chen, Xiang Bai, and Cong Yao, “Textscanner: Reading characters in or- der for robust scene text recognition,” in Proceedings of the AAAI conference on artificial intelligence , 2020, vol. 34, pp. 12120–12127

    2020

  15. [23]

    Seed: Semantics enhanced encoder- decoder framework for scene text recognition,

    Zhi Qiao, Yu Zhou, Dongbao Yang, Yucan Zhou, and Weiping Wang, “Seed: Semantics enhanced encoder- decoder framework for scene text recognition,” in Pro- ceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, 2020, pp. 13528–13537

    2020

  16. [24]

    What if we only use real datasets for scene text recogni- tion? toward scene text recognition with fewer labels,

    Jeonghun Baek, Yusuke Matsui, and Kiyoharu Aizawa, “What if we only use real datasets for scene text recogni- tion? toward scene text recognition with fewer labels,” in Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2021, pp. 3113– 3122

    2021

  17. [25]

    From two to one: A new scene text recognizer with visual language modeling network,

    Yuxin Wang, Hongtao Xie, Shancheng Fang, Jing Wang, Shenggao Zhu, and Yongdong Zhang, “From two to one: A new scene text recognizer with visual language modeling network,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 14194–14203

    2021

  18. [26]

    Levenshtein ocr,

    Cheng Da, Peng Wang, and Cong Yao, “Levenshtein ocr,” in European Conference on Computer Vision . Springer, 2022, pp. 322–338

    2022

  19. [27]

    Multi- modal text recognition networks: Interactive enhance- ments between visual and semantic features,

    Byeonghu Na, Yoonsik Kim, and Sungrae Park, “Multi- modal text recognition networks: Interactive enhance- ments between visual and semantic features,” in Euro- pean Conference on Computer Vision . Springer, 2022, pp. 446–463

    2022

  20. [28]

    Petr: Rethinking the capability of transformer-based language model in scene text recognition,

    Yuxin Wang, Hongtao Xie, Shancheng Fang, Mengting Xing, Jing Wang, Shenggao Zhu, and Yongdong Zhang, “Petr: Rethinking the capability of transformer-based language model in scene text recognition,” IEEE Trans- actions on Image Processing , vol. 31, pp. 5585–5598, 2022

    2022

  21. [29]

    Reading and writing: Discriminative and generative modeling for self-supervised text recognition,

    Mingkun Yang, Minghui Liao, Pu Lu, Jing Wang, Shenggao Zhu, Hualin Luo, Qi Tian, and Xiang Bai, “Reading and writing: Discriminative and generative modeling for self-supervised text recognition,” in Pro- ceedings of the 30th ACM International Conference on Multimedia, 2022, pp...

    2022

  22. [30]

    Clipter: Looking at the bigger picture in scene text recognition,

    Aviad Aberdam, David Bensa ¨ıd, Alona Golts, Roy Ganz, Oren Nuriel, Royee Tichauer, Shai Mazor, and Ron Litman, “Clipter: Looking at the bigger picture in scene text recognition,” arXiv preprint arXiv:2301.07464, 2023

    2023 arXiv

  23. [31]

    Revisiting scene text recognition: A data perspective,

    Qing Jiang, Jiapeng Wang, Dezhi Peng, Chongyu Liu, and Lianwen Jin, “Revisiting scene text recognition: A data perspective,” in Proceedings of the IEEE/CVF international conference on computer vision , 2023, pp. 20543–20554

    2023

  24. [32]

    Synthetic data and artificial neu- ral networks for natural scene text recognition,

    Max Jaderberg, Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman, “Synthetic data and artificial neu- ral networks for natural scene text recognition,” arXiv preprint arXiv:1406.2227, 2014

    2014 arXiv

  25. [33]

    Synthetic data for text localisation in natural images,

    Ankush Gupta, Andrea Vedaldi, and Andrew Zisserman, “Synthetic data for text localisation in natural images,” in Proceedings of the IEEE conference on computer vi- sion and pattern recognition, 2016, pp. 2315–2324

    2016

  26. [34]

    Coco-text: Dataset and benchmark for text detection and recognition in natural images,

    Andreas Veit, Tomas Matera, Lukas Neumann, Jiri Matas, and Serge Belongie, “Coco-text: Dataset and benchmark for text detection and recognition in natural images,” arXiv preprint arXiv:1601.07140, 2016

    2016 arXiv

  27. [35]

    Icdar2017 competition on reading chi- nese text in the wild (rctw-17),

    Baoguang Shi, Cong Yao, Minghui Liao, Mingkun Yang, Pei Xu, Linyan Cui, Serge Belongie, Shijian Lu, and Xiang Bai, “Icdar2017 competition on reading chi- nese text in the wild (rctw-17),” in 2017 14th iapr inter- national conference on document analysis and recogni- tion (ICDA...

    2017

  28. [36]

    Uber-text: A large-scale dataset for optical character recognition from street-level imagery,

    Ying Zhang, Lionel Gueguen, Ilya Zharkov, Peter Zhang, Keith Seifert, and Ben Kadlec, “Uber-text: A large-scale dataset for optical character recognition from street-level imagery,” in SUNw: Scene Understanding Workshop-CVPR, 2017, vol. 2017, p. 5

    2017

  29. [37]

    Icdar2019 robust reading challenge on arbitrary-shaped text-rrc-art,

    Chee Kheng Chng, Yuliang Liu, Yipeng Sun, Chun Chet Ng, Canjie Luo, Zihan Ni, ChuanMing Fang, Shuaitao Zhang, Junyu Han, Errui Ding, et al., “Icdar2019 robust reading challenge on arbitrary-shaped text-rrc-art,” in 2019 International Conference on Document Analysis and Recogni...

    2019

  30. [38]

    Icdar 2019 competition on large-scale street view text with partial labeling-rrc-lsvt,

    Yipeng Sun, Zihan Ni, Chee-Kheng Chng, Yuliang Liu, Canjie Luo, Chun Chet Ng, Junyu Han, Errui Ding, Jingtuo Liu, Dimosthenis Karatzas, et al., “Icdar 2019 competition on large-scale street view text with partial labeling-rrc-lsvt,” in 2019 International Conference on Document...

    2019

  31. [39]

    Icdar2019 robust reading challenge on multi-lingual scene text detection and recognition—rrc- mlt-2019,

    Nibal Nayef, Yash Patel, Michal Busta, Pinaki Nath Chowdhury, Dimosthenis Karatzas, Wafa Khlif, Jiri Matas, Umapada Pal, Jean-Christophe Burie, Cheng- lin Liu, et al., “Icdar2019 robust reading challenge on multi-lingual scene text detection and recognition—rrc- mlt-2019,” in ...

    2019

  32. [40]

    Icdar 2019 robust read- ing challenge on reading chinese text on signboard,

    Rui Zhang, Yongsheng Zhou, Qianyi Jiang, Qi Song, Nan Li, Kai Zhou, Lei Wang, Dong Wang, Minghui Liao, Mingkun Yang, et al., “Icdar 2019 robust read- ing challenge on reading chinese text on signboard,” in 2019 international conference on document analysis and recognition (ICD...

    2019

  33. [41]

    Openimages: A public dataset for large-scale multi- label and multi-class image classification,

    Ivan Krasin, Tom Duerig, Neil Alldrin, Vittorio Ferrari, Sami Abu-El-Haija, Alina Kuznetsova, Hassan Rom, Jasper Uijlings, Stefan Popov, Andreas Veit, et al., “Openimages: A public dataset for large-scale multi- label and multi-class image classification,” Dataset available fr...

    2017

  34. [42]

    Textocr: Towards large-scale end-to-end reasoning for arbitrary-shaped scene text,

    Amanpreet Singh, Guan Pang, Mandy Toh, Jing Huang, Wojciech Galuba, and Tal Hassner, “Textocr: Towards large-scale end-to-end reasoning for arbitrary-shaped scene text,” in Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, 2021, pp. 8802–8812

    2021

  35. [43]

    Open images v5 text annotation and yet another mask text spotter,

    Ilya Krylov, Sergei Nosov, and Vladislav Sovrasov, “Open images v5 text annotation and yet another mask text spotter,” inAsian Conference on Machine Learning. PMLR, 2021, pp. 379–389

    2021

  36. [44]

    Scene text recognition using higher order language priors,

    Anand Mishra, Karteek Alahari, and CV Jawahar, “Scene text recognition using higher order language priors,” in BMVC-British machine vision conference . BMV A, 2012

    2012

  37. [45]

    End- to-end scene text recognition,

    Kai Wang, Boris Babenko, and Serge Belongie, “End- to-end scene text recognition,” in 2011 International conference on computer vision. IEEE, 2011, pp. 1457– 1464

    2011

  38. [46]

    Recognizing text with perspective distortion in natural scenes,

    Trung Quy Phan, Palaiahnakote Shivakumara, Shangx- uan Tian, and Chew Lim Tan, “Recognizing text with perspective distortion in natural scenes,” in Proceed- ings of the IEEE international conference on computer vision, 2013, pp. 569–576

    2013

  39. [47]

    Icdar 2013 robust reading competition,

    Dimosthenis Karatzas, Faisal Shafait, Seiichi Uchida, Masakazu Iwamura, Lluis Gomez i Bigorda, Sergi Rob- les Mestre, Joan Mas, David Fernandez Mota, Jon Al- mazan Almazan, and Lluis Pere De Las Heras, “Icdar 2013 robust reading competition,” in 2013 12th inter- national confe...

    2013

  40. [48]

    Icdar 2015 competition on robust reading,

    Dimosthenis Karatzas, Lluis Gomez-Bigorda, Anguelos Nicolaou, Suman Ghosh, Andrew Bagdanov, Masakazu Iwamura, Jiri Matas, Lukas Neumann, Vijay Ramase- shan Chandrasekhar, Shijian Lu, et al., “Icdar 2015 competition on robust reading,” in 2015 13th interna- tional conference on...

    2015

  41. [49]

    A robust arbi- trary text detection system for natural scene images,

    Anhar Risnumawan, Palaiahankote Shivakumara, Chee Seng Chan, and Chew Lim Tan, “A robust arbi- trary text detection system for natural scene images,” Expert Systems with Applications , vol. 41, no. 18, pp. 8027–8048, 2014

    2014

  42. [50]

    Unrealtext: Synthe- sizing realistic scene text images from the unreal world,

    Shangbang Long and Cong Yao, “Unrealtext: Synthe- sizing realistic scene text images from the unreal world,” arXiv preprint arXiv:2003.10608, 2020

    2003 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.