Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Crop-based Chinese text retrieval fails on vertical, cross-line, and partial layouts; CSTR-CLIP encodes the full image with text-region guidance and multi-granularity alignment, reaching 88.57 mAP on CSVTR, 18.82 points over the prior best.

desk verdict Useful benchmark-and-baseline paper with a serious, unaddressed data-contamination question that should be resolved before the numbers are trusted. read the letter →

arxiv 2506.04999 v1 pith:PYYE5SNK submitted 2025-06-05 cs.CV

Gengluo Li , Huawen Shen , Yu Zhou This is my paper · ORCID
classification cs.CV
keywords ChinesescenetextretrievalDL-CSVTRbenchmarkdiverselayoutsCLIPmulti-granularityalignmenttext-regionsegmentationmaprandomgranularityprocessingcross-modal
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 claims that Chinese scene text retrieval fails in real scenes because it inherited an English-oriented recipe: crop the detected text region, then match it to the query at a single granularity. Chinese text is laid out vertically, spans rows and columns, and embeds query terms inside longer strings, so a crop either cuts the query short or pulls in unrelated characters while discarding the surrounding context. To make the case, the authors contribute DL-CSVTR, a 2,070-image benchmark whose queries appear in vertical, cross-line, and partial layouts, and CSTR-CLIP, a CLIP-based model that encodes the full image steered by a text-region segmentation map and trains first on whole-text alignment, then on randomly varied alignment granularities. They report 88.57 mean average precision (mAP) on the previous CSVTR benchmark, 18.82 points above the best crop-based method, and 84.44, 65.56, and 61.85 mAP on the three new layout subsets, all above every crop-based baseline.

What carries the argument

The load-bearing components are the Text Position Convolution, a single-channel convolution that converts the text segmentation map into a guided embedding added to the CLIP RGB embedding before the image encoder; the Random Alignment Granularity Processing (RAGP) module, which during stage-two training randomly masks characters from the text, merges the nearest neighboring segmentation map, and extends the text toward the nearest box; and per-layer 1x1-convolution fusion layers that inject the frozen original CLIP encoder's global features into the fine-tuned encoder. The segmentation map is a single-channel grayscale image with text pixels highlighted, and the same map used for guidance in training doubles as the user interface for region-specified retrieval at inference.

What would settle it

Build a test set of street-view images in which the text box nearest to each query is deliberately unrelated to it, and compare DL-CSVTR cross-line and partial mAP with RAGP enabled versus disabled; if accuracy drops toward the stage-1 level whenever neighbors are unrelated, the spatial-proximity assumption is the fragile load-bearing premise.

Watch

Extended reading notes

Core claim

The central claim is that scene text retrieval should move from matching cropped text regions to matching a whole image whose attention is steered by text positions. CSTR-CLIP sums the CLIP RGB embedding with a Text Position Convolution that encodes the text segmentation map, runs the result through the CLIP image encoder, and in a second stage adds the frozen original CLIP encoder's global features at every layer plus a random multi-granularity alignment. The authors attribute the CSVTR improvement (69.75 to 88.57 mAP) to retaining visual features outside the text region, and the DL-CSVTR gains on the hardest layouts to RAGP, which deliberately breaks the exact segmentation-map-to-text match during training so that the model perceives text elements inside and near the guided region. On vertical layouts the full-image information itself is credited with most of the gain.

Load-bearing premise

The RAGP augmentation in Section 4 assumes that semantically related text elements are spatially close in the image, so merging a nearby text box into the training pair teaches a useful lesson; when neighboring text is semantically unrelated, the augmented pairs train the model to align a mask with a mismatched string, and the reported cross-line and partial gains would not transfer.

Editorial extensions

If this is right

  • Keeping the whole image instead of cropping detected boxes lifts CSVTR mAP from 69.75 (the best crop-based method) to 88.57, quantifying how much context crop-based retrieval discards.
  • Adding global features and RAGP in stage two raises cross-line mAP from 45.98 to 65.56 and partial mAP from 33.25 to 61.85 on DL-CSVTR, so multi-granularity alignment is the ingredient that handles the hardest layouts.
  • Because the segmentation map is user-controllable, the same model enables region-specified retrieval: highlighting part of the map narrows the search to a chosen area of the image.
  • If the detector finds no text, a fully highlighted segmentation map still lets the model retrieve from visual context alone, something crop-based pipelines cannot do.

Reading between the lines

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

  • The two-stage curriculum of exact alignment followed by deliberately broken alignment could transfer to other vision-language tasks where a region only partially encodes a target concept, such as referring expression grounding.
  • A controlled stress test of RAGP's spatial-proximity assumption, using images where neighboring text boxes are semantically unrelated, would reveal how much of the cross-line and partial gains depend on that assumption; if gains collapse, a language-aware neighbor selector would be the natural fix.
  • The same full-image-plus-guidance design should simplify English scene text retrieval as well, since retrieval accuracy would no longer be gated by whether the detected crop exactly contains the query word.
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

4 major / 6 minor

Summary. The paper addresses Chinese scene text retrieval, arguing that existing benchmarks and methods, inherited from English, fail on diverse Chinese text layouts. It introduces DL-CSVTR, a new test-only benchmark with vertical, cross-line, and partial layout subsets (2,070 images, 89 queries). It also proposes CSTR-CLIP, a two-stage CLIP-based model: stage 1 trains region-focused OCR via segmentation-map-guided single-granularity alignment; stage 2 injects frozen full-image global features and applies a Random Alignment Granularity Processing (RAGP) augmentation that randomly masks, expands, and splices text and segmentation maps to simulate multi-granularity layouts. Experiments report 88.57 mAP on CSVTR (an 18.82-point gain over the prior best) and strong results on DL-CSVTR (84.44/65.56/61.85 mAP on vertical/cross-line/partial), with ablations supporting the contributions of global features and RAGP.

Significance. If the empirical results hold, the paper makes two useful contributions: a benchmark that targets a real gap (Chinese layout diversity in street-view text), and a method that demonstrates the value of moving beyond cropped text regions to full-image features combined with region guidance. The ablations in Table 3 are informative and the authors state that code and data will be released. The main novelty—multi-granularity alignment with global feature injection—is plausible and could influence future scene-text retrieval work. However, the central claims depend on two empirical guarantees that are not yet established: the DL-CSVTR test set must be disjoint from the ReCTS training data, and the RAGP hyperparameters must not be silently selected on the test set. These issues are directly addressable but are load-bearing for the benchmark's validity and the method's reported gains.

major comments (4)
  1. [§3 and §5 (Training Datasets)] The paper does not report any overlap or near-duplicate analysis between the new DL-CSVTR test images and the ReCTS training images. ReCTS is a dataset of real Chinese street-sign photos, and DL-CSVTR is constructed by searching for common signboard phrases (e.g., '封顶大吉', '瑞幸咖啡'); it is entirely plausible that the same or near-duplicate images appear in both. Because ReCTS is used for training in both stages, any overlap would mean the DL-CSVTR results in Table 2 (84.44/65.56/61.85 mAP) partly reflect memorization of test images rather than layout generalization. Please provide a concrete deduplication analysis (e.g., perceptual hashing, image retrieval-based near-duplicate check, or manual inspection) and report the overlap statistics; if overlap exists, remove those images and re-run the evaluation.
  2. [§5 (Implementation Details) and §5.3 (Table 3)] The RAGP hyperparameters β=0.2, α=0.3, and θ=0.2 are stated without any description of a validation protocol or sensitivity analysis. Since RAGP produces large improvements on DL-CSVTR-CL (45.98→60.08) and DL-CSVTR-P (33.25→52.51) in Table 3, the reader cannot rule out that these values were chosen by observing the test-set numbers. Please specify how the hyperparameters were selected (e.g., a held-out validation split from ReCTS or a separate development set) and provide a sensitivity study over these parameters, or the RAGP contribution remains under-validated.
  3. [§4 (RAGP)] The RAGP augmentation explicitly relies on the assumption that 'semantically related text elements are spatially close in the image.' This assumption is load-bearing because RAGP merges each text with the nearest-neighbor text by bounding-box centroid. If nearby text boxes are semantically unrelated, the augmented pairs train the model to align a text string with a mismatched mask, and the reported cross-line/partial gains would likely not transfer. Please provide an empirical test of the assumption—for example, measure the semantic relatedness of the merged pairs in the training set, or compare the current nearest-neighbor rule against a random-neighbor or same-image-random rule to show that the spatial proximity heuristic is actually beneficial.
  4. [§2, §5.1, Table 1] The paper cites Zeng et al. (2024a), a CLIP-based scene text retrieval method that is directly relevant to the proposed approach, but this method is not included as a baseline in Table 1 or Table 2. Given the paper's claim of state-of-the-art performance on CSVTR, omitting a recent, closely related method weakens that claim. Please add this baseline (or justify its exclusion) and report its performance on CSVTR and, if possible, on DL-CSVTR.
minor comments (6)
  1. [Abstract and §5.1] The paper states '18.82% accuracy' when Table 1 reports mAP; please use 'mAP' or 'percentage points' for precision.
  2. [Appendix A.3, Figure 9] Figure 9's caption says 'Distribution of text layouts...' but the figure appears to display the query-term lists for the three DL-CSVTR subsets; this caption seems mismatched with the figure content.
  3. [Appendix A.2] The sentence 'The details of the specific query word settings can be found in the supplementary material' is followed in the appendix by 'as shown in Figure 2'; the query lists actually appear in Figure 9, so the cross-reference appears incorrect.
  4. [§3 (Partial layout)] The phrase 'might include unrelated extra not unrelated characters' appears to be a grammatical error; it should likely read 'extra unrelated characters.'
  5. [Table 3] All results in Table 3 are reported as single runs with no indication of variance; please state whether results are averaged over multiple seeds or runs, or note that only single runs were performed.
  6. [Algorithm 3] In the Random Expand step, the pseudocode says 'Find Tip whose centroid is closest to Tik' but Tik is a text string, not a bounding box; please clarify that the centroid is taken from Tik's bounding box, to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the method is an empirical baseline evaluated on external and newly constructed benchmarks, with no derivation step that reduces to its own inputs.

full rationale

This is an empirical systems paper, not a derivation, and I find no step in which a claimed result is equivalent to its inputs by construction. The model CSTR-CLIP is trained on SynthText-CH and ReCTS, and its stated test sets are CSVTR and the newly introduced DL-CSVTR, which the paper explicitly says is 'used exclusively for testing purposes' (Section 3). The headline improvement over Luo et al. is an externally benchmarked comparison, not a quantity fitted from the same data. The RAGP hyperparameters beta, alpha, and theta are fixed constants and are ablated rather than being defined in terms of the target mAP values; possible selection of hyperparameters on the test set would be a methodological overfitting concern, not circularity. The stated spatial-proximity assumption underlying RAGP is an explicit modeling assumption whose failure would reduce transfer, but it is not a self-referential or fit-derived claim. The paper's self-citations (e.g., Shen et al. 2025, Zeng et al. 2024a) appear only in related-work context and are not load-bearing for the benchmark results or the method's derivation. The skeptic's concern about possible image overlap between DL-CSVTR and ReCTS is a data-contamination and external-validity question; even if true, it would not make the derivation circular. Accordingly, no specific circular step can be exhibited, and the score is 0.

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

The central claims rest on the pretrained Chinese CLIP backbone, the RAGP spatial-closeness heuristic, the use of PaddleOCR detections at inference, and the DL-CSVTR annotation assumptions. The three RAGP probabilities are the only explicit hand-set numbers that directly control the training signal; they are reported as fixed values without a validation sweep.

free parameters (3)
  • RAGP mask probability β = 0.2
    Controls how often characters are masked from the ends of the query text in RAGP; set without a reported validation sweep.
  • RAGP expand probability α = 0.3
    Controls how often the query text is merged with the spatially closest text; chosen by hand.
  • RAGP splice probability θ = 0.2
    Controls how often the segmentation map is concatenated with the closest map; chosen by hand.
assumptions (5)
  • domain assumption Pre-trained Chinese CLIP provides useful shared image-text embeddings and OCR-like capability for Chinese text.
    The model is initialized with Chinese CLIP weights (Yang et al., 2023) and relies on its text-image alignment; no independent evidence in the paper for Chinese OCR ability.
  • domain assumption Semantically related text elements are spatially close in scene images.
    Stated in Section 4 RAGP; used to justify merging nearby text boxes as a training augmentation.
  • domain assumption PaddleOCR detects text lines accurately enough to generate segmentation maps at inference.
    Used in Algorithm 2 for gallery construction; detection errors propagate to the inferred maps.
  • ad hoc to paper The three layout categories (vertical, cross-line, partial) capture the practically important Chinese layout diversity.
    Defined in Section 3; no external survey or agreement statistics justify the categorization.
  • domain assumption Annotator screening ensures uniqueness of layout labels without measured inter-annotator agreement.
    Section 3 describes screening by three annotators but no agreement metric or adjudication rule is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts." pith.science (2026). https://pith.science/paper/PYYE5SNK

@misc{pith2026250604999,
  author       = {Pith},
  title        = {Pith review of: Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PYYE5SNK}},
  note         = {Machine review of arXiv:2506.04999}
}
read the original abstract

Chinese scene text retrieval is a practical task that aims to search for images containing visual instances of a Chinese query text. This task is extremely challenging because Chinese text often features complex and diverse layouts in real-world scenes. Current efforts tend to inherit the solution for English scene text retrieval, failing to achieve satisfactory performance. In this paper, we establish a Diversified Layout benchmark for Chinese Street View Text Retrieval (DL-CSVTR), which is specifically designed to evaluate retrieval performance across various text layouts, including vertical, cross-line, and partial alignments. To address the limitations in existing methods, we propose Chinese Scene Text Retrieval CLIP (CSTR-CLIP), a novel model that integrates global visual information with multi-granularity alignment training. CSTR-CLIP applies a two-stage training process to overcome previous limitations, such as the exclusion of visual features outside the text region and reliance on single-granularity alignment, thereby enabling the model to effectively handle diverse text layouts. Experiments on existing benchmark show that CSTR-CLIP outperforms the previous state-of-the-art model by 18.82% accuracy and also provides faster inference speed. Further analysis on DL-CSVTR confirms the superior performance of CSTR-CLIP in handling various text layouts. The dataset and code will be publicly available to facilitate research in Chinese scene text retrieval.

Figures

Figures reproduced from arXiv: 2506.04999 by the authors.

Figure 1
Figure 1. Pipeline comparison between (a) English scene text re￾trieval, (b) Chinese scene text retrieval adopted in previous work, and (c) Our proposed Chinese scene text retrieval framework that is based on full image information and multi-granularity alignment. The patterns in the circles represent the text layout forms. 1. Introduction Text is an important object in scene images, and scene text related research topics inc… view at source ↗
Figure 2
Figure 2. Layout distribution of the visual instance of the query word in the image in CSVTR. • Our experiments demonstrate that the proposed method not only achieves state-of-the-art performance on the existing CSVTR dataset but also surpasses previous models in retrieval capabilities on the newly introduced DL-CSVTR benchmark. 2. Related Work Scene Text Retrieval Benchmark. Mishra et al. (Mishra et al., 2013) is the first w… view at source ↗
Figure 3
Figure 3. Common Chinese text layouts, where the visual repre￾sentation of query terms is based on the text detection model’s cropped results. In category (d), the red-highlighted area indicates the corresponding query term. line, and partial, as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Two-stage training framework for the CSTR-CLIP model, including: (a) Training of model’s region attention and OCR capabilities based on single-granularity alignment; (b) Adapting CSTR-CLIP to diverse text layouts using random multi-granularity alignment and integrating…
Figure 5
Figure 5. Figure 5: The retrieval pipeline of CSTR-CLIP. the entire text within the guided region, thereby reducing its flexibility. In cross-line and partial layouts, the correspond￾ing text region may miss some queries or include redundant text, making single-granularity alignment less …
Figure 7
Figure 7. Figure 7: An example of region-specified scene text retrieval. The green boxes in the search results indicate visual instances of the query terms. 5.4. Interactive Region-Specified Scene Text Retrieval Our approach paves the way for more diverse text retrieval methods, enabling …
Figure 6
Figure 6. Figure 6: Visualization results of model intermediate layer features, including (a) original CLIP, (b) CSTR-CLIP after Stage 1 training, (c) CSTR-CLIP after Stage 2 training. Query Segmentation Map Top3 Retrival results “蜜雪冰城 ” [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: Distribution of text layouts for visual representations of query terms in CSVTR The statistical results clearly show that the visual layout of most query terms is horizontal, which does not reflect the challenges that diverse text layouts pose to scene text retrieval t…
Figure 9
Figure 9. Figure 9: Distribution of text layouts for visual representations of query terms in CSVTR A.4. Supplementary examples of DL-CSVTR We provide a portion of the DL-CSVTR dataset, ensuring that the images do not reveal information about the author or associated entities. The dataset…
Figure 10
Figure 10. Figure 10: Sample presentation of three benchmarks in DL-CSVTR. Best viewed in zoom. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Visualization of retrieval results. (a) An example on DL-CSVTR-V benchmark, in which rank@1-5 retrieval results are provided. (b) An example on DL-CSVTR-CL benchmark, in which rank@1-5 retrieval results are provided. (c) An example on DL-CSVTR-P benchmark, in which ra…
Figure 12
Figure 12. Figure 12: Comparison of AP of each query word under the DL-CSVTR-V benchmark [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Comparison of AP of each query word under the DL-CSVTR-CL benchmark. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: Comparison of AP of each query word under the DL-CSVTR-P benchmark [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Comparison of AP of each query word under the CSVTR benchmark. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MORE: A Multilingual Document Parsing Benchmark and Evaluation

    cs.CV 2026-07 conditional novelty 6.0 of 10

    MORE provides a 149-language, structure-aware document parsing benchmark from real PDFs and reports baselines showing specialized OCR models still fail on tables and rare scripts.

Reference graph

Works this paper leans on

58 extracted references · 53 canonical work pages · cited by 1 Pith paper

  1. [1]

    Word spotting and recognition with embedded attributes

    Almazán, J., Gordo, A., Fornès, A., and Valveny, E. Word spotting and recognition with embedded attributes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014

    work page 2014

  2. [2]

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

    Bai, X., Yang, M., Lyu, P., Xu, Y., and Luo, J. Integrating scene text and visual appearance for fine-grained image classification. IEEE Access, 2018

    work page 2018

  3. [3]

    Composed image retrieval using contrastive learning and task-oriented clip-based features

    Baldrati, A., Bertini, M., Uricchio, T., and Del Bimbo, A. Composed image retrieval using contrastive learning and task-oriented clip-based features. ACM Transactions on Multimedia Computing, Communications and Applications, 2023

    work page 2023

  4. [4]

    The devil is in fine-tuning and long-tailed problems: A new benchmark for scene text detection

    Cao, T., Lyu, J., Zeng, W., Mu, W., and Zhou, Y. The devil is in fine-tuning and long-tailed problems: A new benchmark for scene text detection. In IJCAI, 2025

    work page 2025

  5. [5]

    Ch'ng, C. K. and Chan, C. S. Total-text: A comprehensive dataset for scene text detection and recognition. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2017

    work page 2017

  6. [6]

    Ghosh, S. K. and Valveny, E. Query by string word spotting based on character bi-gram indexing. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2015

    work page 2015

  7. [7]

    K., Gomez, L., Karatzas, D., and Valveny, E

    Ghosh, S. K., Gomez, L., Karatzas, D., and Valveny, E. Efficient indexing for query by string text retrieval. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2015

    work page 2015

  8. [8]

    Lsde: Levenshtein space deep embedding for query-by-string word spotting

    G \'o mez, L., Rusinol, M., and Karatzas, D. Lsde: Levenshtein space deep embedding for query-by-string word spotting. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2017

    work page 2017

Show all 58 references
  1. [9]

    Single shot scene text retrieval

    G \'o mez, L., Mafla, A., Rusinol, M., and Karatzas, D. Single shot scene text retrieval. In Proceedings of the European Conference on Computer Vision (ECCV), 2018

    2018

  2. [10]

    Synthetic data for text localisation in natural images

    Gupta, A., Vedaldi, A., and Zisserman, A. Synthetic data for text localisation in natural images. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  3. [11]

    Bridging the gap between end-to-end and two-step text spotting

    Huang, M., Li, H., Liu, Y., Bai, X., and Jin, L. Bridging the gap between end-to-end and two-step text spotting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  4. [12]

    Reading text in the wild with convolutional neural networks

    Jaderberg, M., Simonyan, K., Vedaldi, A., and Zisserman, A. Reading text in the wild with convolutional neural networks. International Journal of Computer Vision, 2016

    2016

  5. [13]

    Karaoglu, S., Tao, R., Gevers, T., and Smeulders, A. W. Words matter: Scene text for image classification and retrieval. IEEE Transactions on Multimedia, 2016

    2016

  6. [14]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Li, J., Li, D., Xiong, C., and Hoi, S. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Proceedings of the International Conference on Machine Learning (ICML), 2022

    2022

  7. [15]

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

    Liao, M., Pang, G., Huang, J., Hassner, T., and Bai, X. Mask textspotter v3: Segmentation proposal network for robust scene text spotting. In Proceedings of the European Conference on Computer Vision (ECCV), 2020

    2020

  8. [16]

    J., Wang, B., Li, W., and Shou, M

    Lin, Y., He, C., Wang, A. J., Wang, B., Li, W., and Shou, M. Z. Parrot captions teach clip to spot text, 2024. URL https://arxiv.org/abs/2312.14232

    2024 arXiv

  9. [17]

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

    Liu, Y., Chen, H., Shen, C., He, T., Jin, L., and Wang, L. Abcnet: Real-time scene text spotting with adaptive bezier-curve network. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    2020

  10. [18]

    Clip4clip: An empirical study of clip for end-to-end video clip retrieval and captioning

    Luo, H., Ji, L., Zhong, M., Chen, Y., Lei, W., Duan, N., and Li, T. Clip4clip: An empirical study of clip for end-to-end video clip retrieval and captioning. Neurocomputing, 2022

    2022

  11. [19]

    Visual and semantic guided scene text retrieval

    Luo, H., Ibrayim, M., Hamdulla, A., and Deng, Q. Visual and semantic guided scene text retrieval. The Journal of Supercomputing, 2024

    2024

  12. [20]

    Arbitrary reading order scene text spotter with local semantics guidance

    Lyu, J., Wang, W., Yang, D., Zhong, J., and Zhou, Y. Arbitrary reading order scene text spotter with local semantics guidance. In AAAI, volume 39, pp.\ 5919--5927, 2025 a

    2025

  13. [21]

    TextBlockV2 : Towards precise-detection-free scene text spotting with pre-trained language model

    Lyu, J., Wei, J., Zeng, G., Li, Z., Xie, E., Wang, W., Ma, C., and Zhou, Y. TextBlockV2 : Towards precise-detection-free scene text spotting with pre-trained language model. TOMM, 2025 b

    2025

  14. [22]

    F., Gomez, L., and Karatzas, D

    Mafla, A., Dey, S., Biten, A. F., Gomez, L., and Karatzas, D. Fine-grained image classification and retrieval by combining visual and locally pooled textual features. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2020

    2020

  15. [23]

    Real-time lexicon-free scene text retrieval

    Mafla, A., Tito, R., Dey, S., G \'o mez, L., Rusinol, M., Valveny, E., and Karatzas, D. Real-time lexicon-free scene text retrieval. Pattern Recognition, 2021

    2021

  16. [24]

    Disentangling visual and written concepts in clip

    Materzy \'n ska, J., Torralba, A., and Bau, D. Disentangling visual and written concepts in clip. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022

    2022

  17. [25]

    Word spotting and recognition via a joint deep embedding of image and text

    Mhiri, M., Desrosiers, C., and Cheriet, M. Word spotting and recognition via a joint deep embedding of image and text. Pattern Recognition, 2019

    2019

  18. [26]

    Image retrieval using textual cues

    Mishra, A., Alahari, K., and Jawahar, C. Image retrieval using textual cues. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2013

    2013

  19. [27]

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

    Qiao, L., Tang, S., Cheng, Z., Xu, Y., Niu, Y., Pu, S., and Wu, F. Text perceptron: Towards end-to-end arbitrary-shaped text spotting. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2020 a

    2020

  20. [28]

    SEED : Semantics enhanced encoder-decoder framework for scene text recognition

    Qiao, Z., Zhou, Y., Yang, D., Zhou, Y., and Wang, W. SEED : Semantics enhanced encoder-decoder framework for scene text recognition. In CVPR, pp.\ 13528--13537, 2020 b

    2020

  21. [29]

    PIMNet : A parallel, iterative and mimicking network for scene text recognition

    Qiao, Z., Zhou, Y., Wei, J., Wang, W., Zhang, Y., Jiang, N., Wang, H., and Wang, W. PIMNet : A parallel, iterative and mimicking network for scene text recognition. In ACM MM, pp.\ 2046--2055, 2021

    2021

  22. [30]

    W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al

    Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al. Learning transferable visual models from natural language supervision. In Proceedings of the International Conference on Machine Learning (ICML), 2021

    2021

  23. [31]

    Pic2word: Mapping pictures to words for zero-shot composed image retrieval

    Saito, K., Sohn, K., Zhang, X., Li, C.-L., Lee, C.-Y., Saenko, K., and Pfister, T. Pic2word: Mapping pictures to words for zero-shot composed image retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  24. [32]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural Information Processing Systems, 2022

    2022

  25. [33]

    LDP : Generalizing to multilingual visual information extraction by language decoupled pretraining

    Shen, H., Li, G., Zhong, J., and Zhou, Y. LDP : Generalizing to multilingual visual information extraction by language decoupled pretraining. In AAAI, volume 39, pp.\ 6805--6813, 2025

    2025

  26. [34]

    and Yang, S

    Shi, C. and Yang, S. Logoprompt: Synthetic text images can be good visual prompts for vision-language models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  27. [35]

    What does clip know about a red circle? visual prompt engineering for vlms

    Shtedritski, A., Rupprecht, C., and Vedaldi, A. What does clip know about a red circle? visual prompt engineering for vlms. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023

    2023

  28. [36]

    Perceiving ambiguity and semantics without recognition: An efficient and effective ambiguous scene text detector

    Shu, Y., Wang, W., Zhou, Y., Liu, S., Zhang, A., Yang, D., and Wang, W. Perceiving ambiguity and semantics without recognition: An efficient and effective ambiguous scene text detector. In ACM MM, pp.\ 1851--1862, 2023

    2023

  29. [37]

    Visual text processing: A comprehensive review and unified evaluation

    Shu, Y., Zeng, W., Zhao, F., Chen, Z., Li, Z., Yang, X., Zhou, Y., Rota, P., Bai, X., Jin, L., et al. Visual text processing: A comprehensive review and unified evaluation. arXiv preprint arXiv:2504.21682, 2025

    2025 arXiv

  30. [38]

    Text siamese network for video textual keyframe detection

    Song, H., Wang, H., Huang, S., Xu, P., Huang, S., and Ju, Q. Text siamese network for video textual keyframe detection. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2019

    2019

  31. [39]

    Reclip: A strong zero-shot baseline for referring expression comprehension, 2022

    Subramanian, S., Merrill, W., Darrell, T., Gardner, M., Singh, S., and Rohrbach, A. Reclip: A strong zero-shot baseline for referring expression comprehension, 2022. URL https://arxiv.org/abs/2204.05991

    2022 arXiv

  32. [40]

    Alpha-clip: A clip model focusing on wherever you want

    Sun, Z., Fang, Y., Wu, T., Zhang, P., Zang, Y., Kong, S., Xiong, Y., Lin, D., and Wang, J. Alpha-clip: A clip model focusing on wherever you want. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  33. [41]

    Scene text retrieval via joint text detection and similarity learning

    Wang, H., Bai, X., Yang, M., Zhu, S., Wang, J., and Liu, W. Scene text retrieval via joint text detection and similarity learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 4558--4567, 2021

    2021

  34. [42]

    and Belongie, S

    Wang, K. and Belongie, S. Word spotting in the wild. In Proceedings of the European Conference on Computer Vision (ECCV), 2010

    2010

  35. [43]

    TPSNet : Reverse thinking of thin plate splines for arbitrary shape scene text representation

    Wang, W., Zhou, Y., Lv, J., Wu, D., Zhao, G., Jiang, N., and Wang, W. TPSNet : Reverse thinking of thin plate splines for arbitrary shape scene text representation. In ACM MM, pp.\ 5014--5025, 2022

    2022

  36. [44]

    TextBlock : Towards scene text spotting without fine-grained detection

    Wei, J., Zhang, Y., Zhou, Y., Zeng, G., Qiao, Z., Guo, Y., Wu, H., Wang, H., and Wang, W. TextBlock : Towards scene text spotting without fine-grained detection. In ACM MM, pp.\ 5892--5902, 2022

    2022

  37. [45]

    Visual matching is enough for scene text retrieval

    Wen, L., Wang, Y., Zhang, D., and Chen, G. Visual matching is enough for scene text retrieval. In Proceedings of the Sixteenth ACM International Conference on Web Search and Data Mining (WSDM), 2023

    2023

  38. [46]

    and Brun, A

    Wilkinson, T. and Brun, A. Semantic and verbatim word spotting using deep neural networks. In Proceedings of the International Conference on Frontiers in Handwriting Recognition (ICFHR), 2016

    2016

  39. [47]

    Bridging semantic gaps for language-supervised semantic segmentation

    Xing, Y., Kang, J., Xiao, A., Nie, J., Shao, L., and Lu, S. Bridging semantic gaps for language-supervised semantic segmentation. CoRR, 2023

    2023

  40. [48]

    Chinese clip: Contrastive vision-language pretraining in chinese, 2023

    Yang, A., Pan, J., Lin, J., Men, R., Zhang, Y., Zhou, J., and Zhou, C. Chinese clip: Contrastive vision-language pretraining in chinese, 2023. URL https://arxiv.org/abs/2211.01335

    2023 arXiv

  41. [49]

    IPAD : Iterative, parallel, and diffusion-based network for scene text recognition

    Yang, X., Qiao, Z., and Zhou, Y. IPAD : Iterative, parallel, and diffusion-based network for scene text recognition. IJCV, pp.\ 1--21, 2025

    2025

  42. [50]

    Turning a clip model into a scene text detector

    Yu, W., Liu, Y., Hua, W., Jiang, D., Ren, B., and Bai, X. Turning a clip model into a scene text detector. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  43. [51]

    Beyond OCR + VQA : Towards end-to-end reading and reasoning for robust and accurate textvqa

    Zeng, G., Zhang, Y., Zhou, Y., Yang, X., Jiang, N., Zhao, G., Wang, W., and Yin, X.-C. Beyond OCR + VQA : Towards end-to-end reading and reasoning for robust and accurate textvqa. Pattern Recognition, 138: 0 109337, 2023

    2023

  44. [52]

    Focus, distinguish, and prompt: Unleashing CLIP for efficient and flexible scene text retrieval

    Zeng, G., Zhang, Y., Wei, J., Yang, D., Zhang, P., Gao, Y., Qin, X., and Zhou, Y. Focus, distinguish, and prompt: Unleashing CLIP for efficient and flexible scene text retrieval. In ACM MM, pp.\ 2525--2534, 2024 a

    2024

  45. [53]

    TextCtrl : Diffusion-based scene text editing with prior guidance control

    Zeng, W., Shu, Y., Li, Z., Yang, D., and Zhou, Y. TextCtrl : Diffusion-based scene text editing with prior guidance control. NeurIPS, 37: 0 138569--138594, 2024 b

    2024

  46. [54]

    Icdar 2019 robust reading challenge on reading chinese text on signboard

    Zhang, R., Zhou, Y., Jiang, Q., Song, Q., Li, N., Zhou, K., Wang, L., Wang, D., Liao, M., Yang, M., et al. Icdar 2019 robust reading challenge on reading chinese text on signboard. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), 2019

    2019

  47. [55]

    Linguistics-aware masked image modeling for self-supervised scene text recognition

    Zhang, Y., Liu, C., Wei, J., Yang, X., Zhou, Y., Ma, C., and Ji, X. Linguistics-aware masked image modeling for self-supervised scene text recognition. In CVPR, 2025

    2025

  48. [56]

    C., and Dai, B

    Zhou, C., Loy, C. C., and Dai, B. Extract free dense labels from clip. In Proceedings of the European Conference on Computer Vision (ECCV), 2022 a

    2022

  49. [57]

    C., and Liu, Z

    Zhou, K., Yang, J., Loy, C. C., and Liu, Z. Conditional prompt learning for vision-language models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022 b

    2022

  50. [58]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.