Pith. sign in

REVIEW 4 major objections 6 minor 59 references

Partial Scene Text Retrieval

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that one end-to-end network can retrieve both full text lines and their partial patches from scene images without patch-level annotations, using a ranking MIL training objective and a dynamic-programming partial match…

desk verdict A solid extension that defines partial scene text retrieval and offers a genuinely useful inference-time search, but the RankMIL training premise is under-verified and the benchmark is self-constructed. read the letter →

arxiv 2411.10261 v2 pith:WXFRRJNV submitted 2024-11-15 cs.CV

Hao Wang , Minghui Liao , Zhouyi Xie , Wenyu Liu , Xiang Bai This is my paper · ORCID
classification cs.CV
keywords scenetextretrievalpartialpatchcross-modalsimilaritylearningmultipleinstanceRankMILdynamicprogrammingDPMAarbitrary-shaped
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

The paper is trying to establish that partial scene text retrieval—searching for a snippet of text inside a photographed text line rather than only the whole line—can be solved without any patch-level annotations. It argues that a network trained only on line-level text labels can learn to rank full text lines and localize both contiguous and non-contiguous partial patches in one shared embedding space. If true, it makes text search in images more useful: a query like “BAR” can find the word “BARBEQUE” in a storefront photo and pinpoint the part of the line that matches. The authors report large mAP gains over existing retrieval and spotting methods, especially for the partial-patch task and for Chinese text.

What carries the argument

The central machinery is a training/inference pair rather than a single identity. RankMIL is a ranking multiple-instance-learning loss, $loss(f_{q_j}, f_{p^\theta_r}) = \max(0, -I(t_{q_j} \in T^r(t_{p_i^l}) \land \Delta s >0)(\Delta s - m))$ with $\Delta s = sim_f(f_{q_j}, f_{p^\theta_r}) - sim_f(f_{q_j}, f_{p_i^l})$, which filters noisy bag patches by requiring the best patch to beat the whole-line similarity by margin $m$. DPMA is a dynamic-programming search over a $T \times T$ grid of local cosine similarities, with state transition $S_{x,y} = \max(\{S_{k,y-1}\}_{k=0}^{x}) + s_{x,y}$ (with boundary cases), which selects an ordered subset of the line's local features and thereby outputs a partial-patch feature without enumerating bags. The word-embedding module (characters to a fixed-length sequence via bilinear interpolation and a bidirectional LSTM) and boundary-point text proposals supply the sequential features that make both mechanisms work.

What would settle it

A direct test would be to synthesize text lines with known unequal character widths (e.g., a wide “W” next to a narrow “i”) and query snippets whose ground-truth patch is known; if the bag built by BCA misses the correct patch or RankMIL's ranking premise fails, the model's PPR mAP on such synthetic lines should drop sharply. Concretely, compare RankMIL against a version trained with oracle patch labels on those same lines: if the gap is large for uneven-width Latin text but small for uniform-width Chinese text, the equal-spatial-character assumption is the culprit. A second check would be to inspect DPMA's optimal path on lines with variable widths and see whether the selected local features align with the true characters of the query.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that a text-line-level retrieval model can be extended to partial patches by replacing the standard MIL binary classification with a ranking objective (RankMIL) and by replacing bag construction at inference with a dynamic-programming search (DPMA) over the line's sequential features. The model embeds query strings and detected text-line proposals—represented by boundary points—into a common feature space, with similarity supervised by normalized edit distance. For training, each text line is cut into bags of partial patches with pseudo labels; RankMIL abandons noisy patches whose similarity to the query is below the line-level similarity, and pushes the best patch above it by a margin. At inference, DPMA walks monotonically through a $T \times T$ grid of local similarities between query and line sequence features to select a subset of local features—spatially adjacent or not—that maximizes cumulative similarity, yielding the partial-patch feature directly. The paper reports that this outperforms state-of-the-art approaches by 8.04 and 12.71 mAP on average across three English datasets for line retrieval and patch retrieval, and by 24.45 and 38.06 mAP on Chinese datasets.

Load-bearing premise

The load-bearing premise is that whenever a query string appears in a text line, the bag built from that line always contains some patch whose feature is more similar to the query than the whole line's feature is; this relies on the Bag Constructing Algorithm's assumption that characters occupy equal spatial widths, which the paper admits is “relatively strong and hardly holds” for non-horizontal text.

Editorial extensions

If this is right

  • A single model can rank and localize text lines and their partial patches simultaneously, so image galleries can be searched at the sub-line level without any patch annotations.
  • Because DPMA does not require spatially adjacent features, queries can match non-continuous partial patches—characters separated by other text—which bag-based methods cannot retrieve.
  • Replacing bag construction with DPMA removes the need to enumerate more than 50 patches per line, improving inference speed while raising PPR accuracy (for example, from 70.37 to 75.74 mAP on CSVTRv2).
  • Learning cross-modal similarity directly generalizes to non-Latin scripts better than PHOC or end-to-end recognition, with the largest reported gains on Chinese datasets.
  • The remaining English TIR–PPR gap is attributed by the authors to variable character widths, making proportional fonts the harder case for the equal-width assumption.

Reading between the lines

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

  • If the equal-spatial-character assumption is the real bottleneck, a version of BCA that segments by estimated per-character widths (using attention or character recognition) should close most of the English TIR–PPR gap; this is directly testable with the released code.
  • RankMIL's ranking premise—best patch beats whole line—is a weak-supervision signal that could transfer to other sequence-like objects, such as audio snippets, time series, or video sub-clips, where instance-level labels are missing but sequence-level containment labels exist.
  • DPMA resembles sequence alignment; coupling it with a learned character-width prior might produce a differentiable soft-alignment that makes training and inference consistent, reducing the train/inference mismatch between RankMIL bags and DPMA search.
  • The larger reported gains on Chinese text suggest the method's advantage grows where character widths are uniform; extending it to cursive or proportional scripts may require explicit width modeling.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces partial scene text retrieval (PPR), a task in which a model must retrieve both full text-line instances and continuous/non-continuous partial patches of scene text, given a query string. The proposed PSTR framework trains a shared cross-modal embedding with a text-line retrieval loss (CMSL) and a ranking-based multiple instance learning loss (RankMIL), where pseudo bag labels are generated by a sliding-window Bag Constructing Algorithm (BCA) over ground-truth text-line annotations. At inference, a Dynamic Partial Match Algorithm (DPMA) searches for an optimal subsequence of local features from each detected text-line proposal, avoiding explicit bag construction. The authors report large mAP improvements over prior retrieval and spotting baselines on English (STR, CTR, ArT) and Chinese (CSVTRv2, ReCTS, LSVT) benchmarks, and release code and datasets.

Significance. If the results are reproducible, the paper makes a useful contribution: it is the first to formalize partial scene text retrieval, and its DPMA provides a practical way to search partial patches without dense enumeration. The method works across Latin and non-Latin scripts, and the released code and CSVTRv2 dataset are valuable assets. The ablations are broadly consistent, and the gains over strong spotters such as SwinTextSpotter are substantial. However, the central training mechanism, RankMIL, rests on an unverified premise about bag contents, and several experimental choices (hyperparameter selection, lack of error bars) currently weaken the confidence in the numerical claims.

major comments (4)
  1. [Section 3.4, Eq. (7)] The RankMIL loss is justified by the assertion in Section 3.4 that a positive bag 'always contains a patch that is more similar to the given query text than the text-line instance.' This premise is not established and is in tension with the paper's own statement in Section 3.3.2 that BCA's equal-spatial-character assumption 'is relatively strong and hardly holds,' especially for non-horizontal text. For variable-width characters (e.g., 'W' versus 'i') or curved text, the BCA window can miss the true character span, so the selected patch may be a false positive. When such a patch still has similarity exceeding the line score, Eq. (7) actively reinforces a wrong association; when its similarity does not exceed the line score, the sample is discarded, removing positive supervision exactly on the hard cases where the reported gains are largest. The paper does not report how often the condition Delta_s > 0 actually holds on positive bags, so the training signal's reliability is unknown. I request a direct measurement of this condition before and during training on each dataset (with per-dataset statistics), an analysis of failure cases due to BCA misalignment, and an ablation that shows RankMIL's gain is not an artifact of discarding hard positives. In addition, since BCA labels only contain contiguous substrings, the authors should explain how non-continuous partial patches receive any RankMIL supervision; Table 6 shows large NCPP gains from DPMA, but the training mechanism for NCPP is unspecified.
  2. [Section 4.3.2 and 4.3.3, Tables 3 and Fig. 7] The hyperparameters K (number of boundary point pairs) and m (RankMIL margin) are selected using the same test datasets whose final numbers are reported: K on ArT and LSVT, m on ReCTS. With no validation split, this selection procedure can inflate mAP, particularly for the margin m, which directly controls the RankMIL loss and on which the reported improvements depend. The authors should either reselect these hyperparameters on held-out validation portions (with separate test sets reported) or present sensitivity curves over a grid with standard errors, clearly stating which split was used.
  3. [Tables 4, 5, 7 and 8] All mAP numbers appear to come from a single training run. The central claims, such as the 8.04% and 12.71% average improvements over state-of-the-art methods, would be much more convincing with statistics: at least three independent runs with reported mean and standard deviation, and ideally statistical significance tests for the main comparisons. Without these, it is difficult to judge whether the observed gaps are stable or partly due to stochastic variation, especially because several improvements are in the 2-5% mAP range where run-to-run variance can matter.
  4. [Section 4.1 and Section 4.4] The PPR benchmark is self-constructed: query words are generated by cutting ground-truth substrings, and evaluation is image-level via mAP as in [6]. This is a reasonable design, but the paper does not state how ground-truth positive partial patches are defined for scoring (e.g., whether any detected patch that contains the query substring is considered positive, or whether patch-level overlap is required). The absence of a patch-level metric makes it hard to interpret the PPR mAP numbers, especially for non-continuous patches, and leaves room for the model to rank an image correctly for reasons other than accurate partial-patch localization. Please specify the evaluation protocol precisely and, if feasible, provide a patch-level or localization-based metric to complement image-level mAP.
minor comments (6)
  1. [Algorithm 1] The loop condition 'for i = 1; i < |t| - n; i++' appears to exclude the last valid subsequence; it should likely be 'i <= |t| - n' (or equivalently 'i < |t| - n + 1') to cover all substrings of length n.
  2. [Section 2.2] The sentence 'These MIL-based methods are designed for general objects' is repeated verbatim twice in immediate succession; one copy should be removed.
  3. [Section 3.1.1] The text says 'the parameters T and C are empirically set to 15 and 128'; these are hyperparameters, not learned parameters, and the wording should be corrected.
  4. [Section 3.5, Eq. (8)] The DPMA transition allows a path to move 'one step to the right and at least zero steps down,' but the recurrence with max over previous rows could be misread as permitting zero horizontal progress; please clarify the exact admissible moves and whether the path must touch every query column exactly once.
  5. [Section 4.4.2, Table 8] The citation for ABCNet v2 is inconsistent: the text refers to 'ABCNet v2 [57]', but reference [57] is Mask TextSpotter v3 while ABCNet v2 is [55]. Please correct the citation.
  6. [Figure 7] The caption of Fig. 7 does not state the dataset and split used for the margin sweep; the text says experiments are on ReCTS, but it should be explicit whether this is the test set or a validation split.

Circularity Check

0 steps flagged · score 2.0 of 10

No construction-level circularity: partial-patch labels come from ground-truth strings, and DPMA is an inference-time search; the main caveats are an untested RankMIL premise and test-set hyperparameter tuning, which are correctness risks rather than circular steps.

full rationale

The paper's derivation chain is not circular at the level of definitions or fitted predictions. The TIR similarity targets are normalized edit distances computed from ground-truth text translations (Eqs. 2-4), and the PPR bag labels are generated by BCA from those same ground-truth strings (Section 3.3.1, Algorithm 1), not from the model's own outputs. RankMIL (Eq. 7) uses these external bag labels plus the learned text-line similarity to define a ranking loss; the DPMA (Eq. 8, Algorithm 2) is a monotone-path search over a similarity grid at inference time, so it is not a fitted constant masquerading as a prediction. The paper's own Section 3.3.2 concedes that the equal-spatial-character assumption underlying BCA "is relatively strong and hardly holds", and the Section 3.4 premise that the bag "always contains a patch that is more similar" is asserted rather than verified; these are legitimate correctness risks for variable-width or curved text, but they are not circular because the bag content and similarity targets are independent of the model's output. The self-citations ([13], [39], [56]) are used for the boundary-point representation, the TPGA generation algorithm, and as baselines; none is invoked as a uniqueness theorem or as the sole justification for the central claim, and the paper independently ablates K and m. Tuning m on ReCTS and K on ArT/LSVT before reporting those datasets is an evaluation-protocol weakness, not a construction-level circularity. Overall, the reported gains could fail to generalize, but the derivation does not reduce to its inputs.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several domain assumptions (equal-spacing in bag construction, existence of a better patch in every positive bag, ordered sequence alignment) plus tuned hyperparameters m, K, and T. There are no invented physical entities; the new components are algorithms and a dataset, which are artifacts rather than postulated entities.

free parameters (3)
  • Margin m in RankMIL = 0.2
    Tuned on ReCTS using Fig. 7; controls how much the selected patch similarity must exceed the text-line similarity and which samples are treated as noisy.
  • Boundary point pairs K = 7
    Selected from {3, 5, 7, 9} on ArT and LSVT in Table 3; fixed to 7 for all reported experiments.
  • Sequence length T and feature dimension C = T=15, C=128
    Empirically set in Section 3.1.1; T directly defines the DPMA search grid size.
assumptions (5)
  • domain assumption Each character in a text line occupies approximately equal spatial space, so sliding windows of equal width produce patches aligned to character substrings.
    Used in the Bag Constructing Algorithm to generate bag proposals and pseudo-labels; Section 3.3.2 admits this assumption "is relatively strong and hardly holds", especially for non-horizontal or variable-width text.
  • domain assumption For a positive bag, there always exists at least one sampled patch whose learned feature is more similar to the query feature than the whole text-line feature.
    Stated as an observation in Section 3.4 and used as the basis for the RankMIL loss; the paper provides no formal justification or failure analysis.
  • domain assumption Query words and text-line instances can be represented as ordered sequences of fixed length T features that support monotonic alignment for partial matching.
    Word embeddings are interpolated to T positions and RoIAlign features are resampled; the DPMA assumes a monotone path on a T by T grid in Section 3.5.
  • domain assumption Normalized edit distance is an appropriate similarity target for scene text retrieval.
    Used as the supervision target in Eqs. 2 through 4; this is a standard but nontrivial modeling choice for text matching.
  • standard math Standard deep learning components (FPN, FCOS, BiLSTM, ArbitraryRoIAlign) work as described in the cited papers.
    The method relies on prior architectures and training procedures without re-deriving them.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Partial Scene Text Retrieval." pith.science (2026). https://pith.science/paper/WXFRRJNV

@misc{pith2026241110261,
  author       = {Pith},
  title        = {Pith review of: Partial Scene Text Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WXFRRJNV}},
  note         = {Machine review of arXiv:2411.10261}
}
read the original abstract

The task of partial scene text retrieval involves localizing and searching for text instances that are the same or similar to a given query text from an image gallery. However, existing methods can only handle text-line instances, leaving the problem of searching for partial patches within these text-line instances unsolved due to a lack of patch annotations in the training data. To address this issue, we propose a network that can simultaneously retrieve both text-line instances and their partial patches. Our method embeds the two types of data (query text and scene text instances) into a shared feature space and measures their cross-modal similarities. To handle partial patches, our proposed approach adopts a Multiple Instance Learning (MIL) approach to learn their similarities with query text, without requiring extra annotations. However, constructing bags, which is a standard step of conventional MIL approaches, can introduce numerous noisy samples for training, and lower inference speed. To address this issue, we propose a Ranking MIL (RankMIL) approach to adaptively filter those noisy samples. Additionally, we present a Dynamic Partial Match Algorithm (DPMA) that can directly search for the target partial patch from a text-line instance during the inference stage, without requiring bags. This greatly improves the search efficiency and the performance of retrieving partial patches. The source code and dataset are available at https://github.com/lanfeng4659/PSTR.

Figures

Figures reproduced from arXiv: 2411.10261 by the authors.

Figure 1
Figure 1. Result examples of retrieving scene text instances. The target text [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The training phase of our proposed framework. Given an image, text-line proposals are detected, and a bag is constructed within text-line [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The representation of boundary points (points in green or red) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: An example of constructing a bag (c) from the labeled text-line [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: The illustration of the DPMA. (a): Features [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: The influence of the parameter m in the RankMIL to the average performances of the TIR and PPR tasks. at the 40kth iteration. The fine-tuning process is terminated at the 80kth iteration. For testing, the longer sides of input images are resized to 960 while the aspect…
Figure 8
Figure 8. Figure 8: The reason why “Baseline + Bags” and “Baseline + DPMA” can [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Comparison of retrieving non-continuous partial patches by the DPMA and constructing bags. (a): detected text-line proposals; (b): retrieved [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Visualization of the retrieval results. The three-row examples are from the results of TIR, PPR (CPP), and PPR (NCPP), respectively. [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Visualization of the retrieval results. The three-row examples are from the results of TIR, PPR (CPP), and PPR (NCPP), respectively. The [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 55 canonical work pages

  1. [6]

    Single shot scene text retrieval,

    L. G ´omez, A. Mafla, M. Rusi ˜nol, and D. Karatzas, “Single shot scene text retrieval,” in ECCV, 2018

    work page 2018

  2. [1]

    Image retrieval using textual cues,

    A. Mishra, K. Alahari, and C. V . Jawahar, “Image retrieval using textual cues,” in ICCV, 2013

    work page 2013

  3. [2]

    Smart library: Identifying books on library shelves using super- vised deep learning for scene text reading,

    X. Yang, D. He, W. Huang, A. Ororbia, Z. Zhou, D. Kifer, and C. L. Giles, “Smart library: Identifying books on library shelves using super- vised deep learning for scene text reading,” in JCDL, 2017

    work page 2017

  4. [3]

    Text siamese network for video textual key frame detection,

    H. Song, H. Wang, S. Huang, P. Xu, S. Huang, and Q. Ju, “Text siamese network for video textual key frame detection,” in ICDAR, 2019

    work page 2019

  5. [4]

    Integrating scene text and visual appearance for fine-grained image classification,

    X. Bai, M. Yang, P. Lyu, Y . Xu, and J. Luo, “Integrating scene text and visual appearance for fine-grained image classification,” IEEE Access , vol. 6, pp. 66 322–66 335, 2018

    work page 2018

  6. [5]

    Knowledge mining with scene text for fine-grained recognition,

    H. Wang, J. Liao, T. Cheng, Z. Gao, H. Liu, B. Ren, X. Bai, and W. Liu, “Knowledge mining with scene text for fine-grained recognition,” in CVPR, 2022

    work page 2022

  7. [7]

    Real-time lexicon-free scene text retrieval,

    A. Mafla, R. Tito, S. Dey, L. G ´omez, M. Rusi ˜nol, E. Valveny, and D. Karatzas, “Real-time lexicon-free scene text retrieval,”Pattern Recog- nition, vol. 110, p. 107656, 2020

    work page 2020

  8. [8]

    Visual matching is enough for scene text retrieval,

    L. Wen, Y . Wang, D. Zhang, and G. Chen, “Visual matching is enough for scene text retrieval,” in ACM WSDM, 2023

    work page 2023

Show all 59 references
  1. [9]

    Word spotting and recognition with embedded attributes,

    J. Almaz ´an, A. Gordo, A. Forn ´es, and E. Valveny, “Word spotting and recognition with embedded attributes,” TPAMI, vol. 36, no. 12, pp. 2552– 2566, 2014

    2014

  2. [10]

    Reading text in the wild with convolutional neural networks,

    M. Jaderberg, K. Simonyan, A. Vedaldi, and A. Zisserman, “Reading text in the wild with convolutional neural networks,” IJCV, vol. 116, no. 1, pp. 1–20, 2016

    2016

  3. [11]

    An end-to-end text spotter with explicit alignment and attention,

    T. He, Z. Tian, W. Huang, C. Shen, Y . Qiao, and C. Sun, “An end-to-end text spotter with explicit alignment and attention,” in CVPR, 2018

    2018

  4. [12]

    Multiple instance boosting for object detection,

    P. A. Viola, J. C. Platt, and C. Zhang, “Multiple instance boosting for object detection,” in NIPS, 2005

    2005

  5. [13]

    Scene text retrieval via joint text detection and similarity learning,

    H. Wang, X. Bai, M. Yang, S. Zhu, J. Wang, and W. Liu, “Scene text retrieval via joint text detection and similarity learning,” in CVPR, 2021

    2021

  6. [14]

    Integrating visual and textual cues for query-by-string word spotting,

    D. Aldavert, M. Rusi ˜nol, R. Toledo, and J. Llad ´os, “Integrating visual and textual cues for query-by-string word spotting,” in ICDAR, 2013

    2013

  7. [15]

    Phocnet: A deep convolutional neural network for word spotting in handwritten documents,

    S. Sudholt and G. A. Fink, “Phocnet: A deep convolutional neural network for word spotting in handwritten documents,” in ICFHR, 2016

    2016

  8. [16]

    Semantic and verbatim word spotting using deep neural networks,

    T. Wilkinson and A. Brun, “Semantic and verbatim word spotting using deep neural networks,” in ICFHR, 2016

    2016

  9. [17]

    LSDE: levenshtein space deep embedding for query-by-string word spotting,

    L. G ´omez, M. Rusi ˜nol, and D. Karatzas, “LSDE: levenshtein space deep embedding for query-by-string word spotting,” in ICDAR, 2017

    2017

  10. [18]

    Binary codes capable of correcting deletions, insertions, and reversals,

    V . Levenshtein, “Binary codes capable of correcting deletions, insertions, and reversals,” in Soviet physics doklady , vol. 10, no. 8, 1966, pp. 707– 710

    1966

  11. [19]

    Efficient indexing for query by string text retrieval,

    S. K. Ghosh, L. G ´omez, D. Karatzas, and E. Valveny, “Efficient indexing for query by string text retrieval,” in ICDAR, 2015

    2015

  12. [20]

    Solving the multiple instance problem with axis-parallel rectangles,

    T. G. Dietterich, R. H. Lathrop, and T. Lozano-P ´erez, “Solving the multiple instance problem with axis-parallel rectangles,” Artif. Intell. , vol. 89, no. 1-2, pp. 31–71, 1997

    1997

  13. [21]

    Harvesting mid-level visual concepts from large-scale internet images,

    Q. Li, J. Wu, and Z. Tu, “Harvesting mid-level visual concepts from large-scale internet images,” in CVPR, 2013

    2013

  14. [22]

    Learning multi-instance deep discriminative patterns for image classification,

    P. Tang, X. Wang, B. Feng, and W. Liu, “Learning multi-instance deep discriminative patterns for image classification,”IEEE TIP ., vol. 26, no. 7, pp. 3385–3396, 2017

    2017

  15. [23]

    Fully convolutional multi-class multiple instance learning,

    D. Pathak, E. Shelhamer, J. Long, and T. Darrell, “Fully convolutional multi-class multiple instance learning,” inICLR, Y . Bengio and Y . LeCun, Eds., 2015

    2015

  16. [24]

    Is object localization for free? - weakly-supervised learning with convolutional neural networks,

    M. Oquab, L. Bottou, I. Laptev, and J. Sivic, “Is object localization for free? - weakly-supervised learning with convolutional neural networks,” in CVPR, 2015. 14

    2015

  17. [25]

    Textboxes: A fast text detector with a single deep neural network,

    M. Liao, B. Shi, X. Bai, X. Wang, and W. Liu, “Textboxes: A fast text detector with a single deep neural network,” in AAAI, S. P. Singh and S. Markovitch, Eds., 2017

    2017

  18. [26]

    Detecting text in natural image with connectionist text proposal network,

    Z. Tian, W. Huang, T. He, P. He, and Y . Qiao, “Detecting text in natural image with connectionist text proposal network,” in ECCV, B. Leibe, J. Matas, N. Sebe, and M. Welling, Eds., 2016

    2016

  19. [27]

    EAST: an efficient and accurate scene text detector,

    X. Zhou, C. Yao, H. Wen, Y . Wang, S. Zhou, W. He, and J. Liang, “EAST: an efficient and accurate scene text detector,” in CVPR, 2017

    2017

  20. [28]

    MOST: A multi-oriented scene text detector with localization refinement,

    M. He, M. Liao, Z. Yang, H. Zhong, J. Tang, W. Cheng, C. Yao, Y . Wang, and X. Bai, “MOST: A multi-oriented scene text detector with localization refinement,” in CVPR, 2021

    2021

  21. [29]

    Textboxes++: A single-shot oriented scene text detector,

    M. Liao, B. Shi, and X. Bai, “Textboxes++: A single-shot oriented scene text detector,” IEEE TIP ., vol. 27, no. 8, pp. 3676–3690, 2018

    2018

  22. [30]

    Rotation-sensitive regression for oriented scene text detection,

    M. Liao, Z. Zhu, B. Shi, G. Xia, and X. Bai, “Rotation-sensitive regression for oriented scene text detection,” in CVPR, 2018

    2018

  23. [31]

    Shape robust text detection with progressive scale expansion network,

    W. Wang, E. Xie, X. Li, W. Hou, T. Lu, G. Yu, and S. Shao, “Shape robust text detection with progressive scale expansion network,” inCVPR, 2019

    2019

  24. [32]

    Real-time scene text detection with differentiable binarization

    M. Liao, Z. Wan, C. Yao, K. Chen, and X. Bai, “Real-time scene text detection with differentiable binarization.” in AAAI, 2020

    2020

  25. [33]

    Real-time scene text detection with differentiable binarization and adaptive scale fusion,

    M. Liao, Z. Zou, Z. Wan, C. Yao, and X. Bai, “Real-time scene text detection with differentiable binarization and adaptive scale fusion,” IEEE TPAMI, vol. 45, no. 1, pp. 919–931, 2022

    2022

  26. [34]

    Towards uncon- strained end-to-end text spotting,

    S. Qin, A. Bissacco, M. Raptis, Y . Fujii, and Y . Xiao, “Towards uncon- strained end-to-end text spotting,” in ICCV, 2019

    2019

  27. [35]

    Mask textspotter: An end-to-end trainable neural network for spotting text with arbitrary shapes,

    M. Liao, P. Lyu, M. He, C. Yao, W. Wu, and X. Bai, “Mask textspotter: An end-to-end trainable neural network for spotting text with arbitrary shapes,” TPAMI, 2019

    2019

  28. [36]

    Text perceptron: Towards end-to-end arbitrary-shaped text spotting,

    L. Qiao, S. Tang, Z. Cheng, Y . Xu, Y . Niu, S. Pu, and F. Wu, “Text perceptron: Towards end-to-end arbitrary-shaped text spotting,” in AAAI, 2020

    2020

  29. [37]

    Character region attention for text spotting,

    Y . Baek, S. Shin, J. Baek, S. Park, J. Lee, D. Nam, and H. Lee, “Character region attention for text spotting,” in ECCV, A. Vedaldi, H. Bischof, T. Brox, and J. Frahm, Eds., vol. 12374, 2020

    2020

  30. [38]

    Abcnet: Real-time scene text spotting with adaptive bezier-curve network,

    Y . Liu, H. Chen, C. Shen, T. He, L. Jin, and L. Wang, “Abcnet: Real-time scene text spotting with adaptive bezier-curve network,” in CVPR, 2020

    2020

  31. [39]

    All you need is boundary: Toward arbitrary-shaped text spotting,

    H. Wang, P. Lu, H. Zhang, M. Yang, X. Bai, Y . Xu, M. He, Y . Wang, and W. Liu, “All you need is boundary: Toward arbitrary-shaped text spotting,” in AAAI, 2020

    2020

  32. [40]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in CVPR, 2017

    2017

  33. [41]

    Glove: Global vectors for word representation,

    J. Pennington, R. Socher, and C. D. Manning, “Glove: Global vectors for word representation,” in ACL, 2014

    2014

  34. [42]

    Efficient estimation of word representations in vector space,

    T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient estimation of word representations in vector space,” in ICLR, 2013

    2013

  35. [43]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  36. [44]

    Fcos: Fully convolutional one- stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: Fully convolutional one- stage object detection,” in ICCV, 2019

    2019

  37. [45]

    Training region-based object detectors with online hard example mining,

    A. Shrivastava, A. Gupta, and R. B. Girshick, “Training region-based object detectors with online hard example mining,” in CVPR, 2016, pp. 761–769

    2016

  38. [46]

    Punctuation as implicit annotations for chinese word segmentation,

    Z. Li and M. Sun, “Punctuation as implicit annotations for chinese word segmentation,” Comput. Linguistics, vol. 35, no. 4, pp. 505–512, 2009

    2009

  39. [47]

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

    A. Veit, T. Matera, L. Neumann, J. Matas, and S. Belongie, “Coco- text: Dataset and benchmark for text detection and recognition in natural images,” arXiv preprint arXiv:1601.07140, 2016

    2016 arXiv

  40. [48]

    ICDAR2019 robust reading challenge on arbitrary-shaped text - rrc-art,

    C. K. Chng, E. Ding, J. Liu, D. Karatzas, C. S. Chan, L. Jin, Y . Liu, Y . Sun, C. C. Ng, C. Luo, Z. Ni, C. Fang, S. Zhang, and J. Han, “ICDAR2019 robust reading challenge on arbitrary-shaped text - rrc-art,” in ICDAR, 2019, pp. 1571–1576

    2019

  41. [49]

    ICDAR 2019 robust reading challenge on reading chinese text on signboard,

    R. Zhang, M. Yang, X. Bai, B. Shi, D. Karatzas, S. Lu, C. V . Jawahar, Y . Zhou, Q. Jiang, Q. Song, N. Li, K. Zhou, L. Wang, D. Wang, and M. Liao, “ICDAR 2019 robust reading challenge on reading chinese text on signboard,” in ICDAR, 2019, pp. 1577–1581

    2019

  42. [50]

    ICDAR 2019 competition on large-scale street view text with partial labeling - RRC-LSVT,

    Y . Sun, D. Karatzas, C. S. Chan, L. Jin, Z. Ni, C. K. Chng, Y . Liu, C. Luo, C. C. Ng, J. Han, E. Ding, and J. Liu, “ICDAR 2019 competition on large-scale street view text with partial labeling - RRC-LSVT,” in ICDAR, 2019

    2019

  43. [51]

    Total-text: A comprehensive dataset for scene text detection and recognition,

    C. K. Ch’ng and C. S. Chan, “Total-text: A comprehensive dataset for scene text detection and recognition,” in ICDAR, 2017

    2017

  44. [52]

    Curved scene text detection via transverse and longitudinal sequence connection,

    Y . Liu, L. Jin, S. Zhang, C. Luo, and S. Zhang, “Curved scene text detection via transverse and longitudinal sequence connection,” Pattern Recognit., vol. 90, pp. 337–345, 2019

    2019

  45. [53]

    Synthetic data for text localisation in natural images,

    A. Gupta, A. Vedaldi, and A. Zisserman, “Synthetic data for text localisation in natural images,” in CVPR, 2016

    2016

  46. [54]

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

    N. Nayef, Y . Patel, M. Busta, P. N. Chowdhury, D. Karatzas, W. Khlif, J. Matas, U. Pal, J.-C. Burie, C.-l. Liu et al. , “Icdar2019 robust reading challenge on multi-lingual scene text detection and recognition—rrc-mlt- 2019,” in ICDAR, 2019

    2019

  47. [55]

    Abcnet v2: Adaptive bezier-curve network for real-time end-to-end text spotting,

    Y . Liu, C. Shen, L. Jin, T. He, P. Chen, C. Liu, and H. Chen, “Abcnet v2: Adaptive bezier-curve network for real-time end-to-end text spotting,” IEEE TPAMI, vol. 44, no. 11, pp. 8048–8064, 2021

    2021

  48. [56]

    Boundary textspotter: Toward arbitrary-shaped scene text spotting,

    P. Lu, H. Wang, S. Zhu, J. Wang, X. Bai, and W. Liu, “Boundary textspotter: Toward arbitrary-shaped scene text spotting,” IEEE TIP , vol. 31, pp. 6200–6212, 2022

    2022

  49. [57]

    Mask textspotter v3: Segmentation proposal network for robust scene text spotting,

    M. Liao, G. Pang, J. Huang, T. Hassner, and X. Bai, “Mask textspotter v3: Segmentation proposal network for robust scene text spotting,” in ECCV, 2020

    2020

  50. [58]

    Swintextspotter: Scene text spotting via better synergy between text detection and text recognition,

    M. Huang, Y . Liu, Z. Peng, C. Liu, D. Lin, S. Zhu, N. Yuan, K. Ding, and L. Jin, “Swintextspotter: Scene text spotting via better synergy between text detection and text recognition,” in CVPR, 2022

    2022

  51. [59]

    Swin transformer: Hierarchical vision transformer using shifted win- dows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted win- dows,” in ICCV, 2021

    2021

Pith tools

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