Pith. sign in

REVIEW 4 major objections 4 minor 40 references

Large-scale Tag-based Font Retrieval with Generative Feature Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read A staged recognition-retrieval model that learns font features by reconstructing synthetic glyphs in different characters achieves 28.08 mAP on single-tag queries, nearly doubling the best prior retrieval baseline.

desk verdict A solid, useful font-retrieval paper with a genuinely new dataset and a clever use of synthetic fonts, but the headline claim of state-of-the-art superiority is under-supported because the main evaluation uses noisy web tags and the human evaluation never compares against the baselines. read the letter →

arxiv 1909.02072 v2 pith:P2FGERC2 submitted 2019-09-04 cs.CV

classification cs.CV
keywords fontretrievaltag-basedgenerativefeaturelearningattentionmechanismtaggingdatasetrecognition-retrievalmodelmulti-labelrecognitionconditionalGAN
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 tries to establish that large-scale tag-based font retrieval can be solved by a staged model that first recognizes tags from glyph images and then ranks fonts, rather than by learning a joint image-text embedding. To make the visual feature capture font identity rather than letter shapes, it exploits the fact that font images are synthetic: a GAN must reconstruct a same-font different-character image from the feature, stripping away character content. An attention map derived from predicted font class re-weights the feature, and a retrieval head maps tag probabilities to affinity scores for multi-tag queries. On a newly collected benchmark of nearly 20,000 professional fonts and 1,824 tags, the full model nearly doubles the best baseline's single-tag mAP and improves human-ranked accuracy.

What carries the argument

The machinery has three parts. First, a generative feature learning module: because font glyphs can be rendered on demand, the network is given a standard-font glyph and must generate, from the hidden feature of an input glyph, a new glyph of the same font in a different character; a PatchGAN discriminator plus L1 reconstruction loss forces the feature to encode font design while discarding character shape. Second, an attentive feature selection module: a font classification head predicts a softmax distribution over font identities, and a fully-connected layer with sigmoid turns that distribution into a node-level attention map that re-weights the hidden feature before tag prediction, with training-time aggregation over several glyphs of the same font. Third, a retrieval model: a two-layer network maps the element-wise product of predicted tag probabilities and a binary query vector, through a power activation, to an affinity score, trained with a pairwise soft-ranking loss on multi-tag triplets.

What would settle it

A reader could take the released dataset, re-label a held-out subset of fonts with professional typographers using the same tag vocabulary, and rerun the MyFonts-test protocol on only those fonts; if the full model's margin over the best retrieval baseline collapses on the clean labels, the reported improvement was an artifact of community-label noise rather than of the generative feature and attention machinery.

Watch

Extended reading notes

Core claim

The paper's central claim is that tag-based font retrieval should be staged as recognition followed by ranking, and that the visual features for that pipeline should be made character-invariant by exploiting the synthetic nature of font images. On the MyFonts-test benchmark the full model achieves 28.08 mAP on single-tag queries and 16.74 mAP on multi-tag queries, against 15.33 and 7.52 for the best adapted retrieval baseline; on the AMT human-ranked set it selects the agreed ground-truth font in 47.50% of groups, up from 44.49% for the basic recognition model. The ablation shows each added component—generative feature learning, attentive feature selection, and the retrieval head—improves results monotonically. The paper takes this as evidence that the recognition-first design, the cross-character reconstruction objective, and the class-prediction attention map together capture font semantics that plain joint-embedding retrieval misses.

Load-bearing premise

The load-bearing premise is that the MyFonts community tag lists, after normalization, are accurate enough as ground truth that training and test metrics reflect real font semantics; if those labels are systematically noisy or skewed, the reported gains may not transfer to what users actually mean by a tag.

Editorial extensions

If this is right

  • A recognition model trained only with multi-label cross-entropy already beats joint-embedding retrieval baselines on this task, so font-tag semantics may be better captured by per-tag classifiers than by a shared image-text embedding space.
  • The generative cross-character reconstruction visibly removes character traces from features: retrieval improves with the GAN module even though the basic recognition model is already strong.
  • The attention-from-font-class design lets the model emphasize different glyph features for different fonts, matching the observation that one visual dimension can tag some fonts and be irrelevant for others.
  • The retrieval head addresses the tag-dominance problem for multi-tag queries, letting rare tags such as 'noisy' participate in ranking alongside frequent tags such as 'script'.
  • Human-ranked accuracy on the AMT set rises with the full model, indicating that the learned ranking orders fonts more consistently with human preferences.

Reading between the lines

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

  • The same render-and-reconstruct-across-a-nuisance-factor trick should transfer to other domains where images are synthetic or renderable, such as icons, CAD parts, signboards, or maps, where an algorithm can deliberately vary one factor to make features invariant to it.
  • Because the MyFonts labels are community-contributed, the model likely learns tag patterns that include foundries' marketing language; a cleaner expert-annotated tag set would test whether the semantic ranking generalizes beyond the benchmark's label distribution.
  • The attention-from-predicted-class idea could be applied to other fine-grained recognition tasks where the class distribution is informative about which visual attributes matter, such as bird or car attribute retrieval.
  • A zero-shot extension is a natural next step: with tag embeddings or a font-tag ontology, the recognition head could produce probabilities for tags never seen in training, which the retrieval head could then rank.
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 / 4 minor

Summary. The paper addresses large-scale tag-based font retrieval. It introduces a dataset of 18,815 fonts from MyFonts with 1,824 tags and a human-ranked AMT evaluation set. The proposed method trains a multi-label tag recognition model, enhances the visual features with a generative module that reconstructs glyph images of different characters, applies implicit attention based on font class predictions, and adds a relation-network-style retrieval model for multi-tag queries. Experiments on MyFonts-test report large mAP/nDCG gains over GNA-RNN, DeViSE, and RelationNet, and an ablation shows additive improvements. On the AMT-test the full model outperforms its own basic recognition version.

Significance. If the results hold, the paper makes a strong contribution: it is, to my knowledge, the first large-scale benchmark dataset for tag-based font retrieval, and the generative feature learning idea leverages the synthetic nature of font images in a principled way. The proposed method shows large, consistent gains over adapted baselines on the MyFonts-test set, and the ablation isolates the contribution of each module. The dataset release and the AMT evaluation protocol are valuable resources. However, the claim of 'significantly outperforms the state-of-the-art' rests on MyFonts-test, whose labels are acknowledged to be noisy and incomplete, while the only human-judgment comparison is against the paper's own basic model. These gaps reduce the strength of the headline claim.

major comments (4)
  1. [Section 3, Section 5.3, Table 1] The MyFonts-test ground truth is derived from the same web-sourced MyFonts tags that define the training labels in Eq. (1). The paper itself states (Section 3) that these tags are 'inevitably noisy and incomplete' and that 'some tags may be omitted by web users for a given font' (Section 5.3). Under the multi-label cross-entropy loss of Eq. (1), an omitted tag is treated as a negative even when the font visually matches, and the same incomplete labels are used as queries and relevance judgments on MyFonts-test. Consequently, the reported mAP advantage (28.08 vs 15.33 against RelationNet) may reflect, to an unknown degree, recovery of MyFonts foundry tagging conventions rather than true semantic font-tag relevance. The AMT-test set is the designed guard against this, but Table 2 compares only the full model with its own basic recognition version; none of the baselines (RelationNet, GNA-RNN, DeViSE) is evaluated on AMT-test. To support the abstract's claim of significant improvement over the state of the art, the authors should report AMT-test accuracy and average rank for the baselines, or otherwise justify why MyFonts-test is a sufficient criterion.
  2. [Table 1, Section 5.2, Eq. (7)] The paper reports no error bars, confidence intervals, or significance tests for any metric. The margins over the baselines are large, but the ablation increments are small (e.g., from 27.75 to 28.08 mAP when adding the generative module on top of attention, and from 16.06 to 16.74 mAP on multi-tag queries when adding the retrieval model). Without a variance estimate or a significance test, the claim that each component 'gets steady improvements' is not statistically supported. I request at least three runs with standard deviations, or bootstrap intervals, for the main configurations.
  3. [Section 4.4] The sentence 'It shows better performance in transforming probabilities than other common activation functions' is not supported by any experiment in the paper. Please provide an ablation comparing the power activation x->(x+epsilon)^alpha with alternatives (e.g., identity, sigmoid, ReLU) or remove this claim.
  4. [Section 3, Section 5.4] The AMT evaluation set is constructed by keeping only groups in which all three workers agree (1,661 groups out of the 9,000 generated). This unanimity filter likely selects easier groups, and the paper reports neither the number of discarded groups nor the distribution of agreements. Because the full model and the basic model are compared on the same retained groups, the internal comparison is fair, but this selection makes the absolute accuracy values hard to interpret, and any future baseline comparison on AMT-test should use the same retained groups. Please report the worker-agreement statistics and consider a supplementary evaluation on all groups with a majority-vote label.
minor comments (4)
  1. [Figure 5] Figure 5 includes a 'Shared Font Class FC Layer' that is not described in the text; clarify its role or remove it from the schematic.
  2. [Section 5.2] The hyper-parameter settings (lambda=10, beta=0.04, J=4, alpha=0.1, gamma=100) are given only for the final model; a sensitivity analysis of alpha and gamma would clarify the robustness of the retrieval model.
  3. [Supplementary Section 8.3] The attention module weights are initialized with a normal distribution of mu=0, sigma=5; this is an unusually large variance and may be a typo (possibly 0.5). Please verify.
  4. [Section 5.3] The paper states that the training data may be incomplete but provides no quantification of tag coverage or noise rates; an analysis of tag frequency and per-font tag counts would help readers assess the label quality.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main results are evaluated on held-out test fonts and on an externally collected AMT human ranking set, and the core training objectives do not reduce to the measured quantities.

full rationale

The paper's central claims are supported by held-out evaluation, not by construction. The tag recognition model (Section 4.1, Eq. 1) is trained with cross-entropy on MyFonts tags, but MyFonts-test uses 1,877 held-out fonts and the AMT-test uses 1,661 human-ranked groups collected from Mechanical Turk, so the reported mAP and accuracy are not fitted to the test labels. The generative feature learning module (Section 4.2, Eqs. 2-5) uses a GAN and L1 reconstruction objective on raw glyph images, which is self-supervised with respect to tags and does not encode the tag labels into the feature. The attentive feature selection module (Section 4.3) is trained using font identity classification, not the tag labels being predicted, and the final retrieval model (Section 4.4, Eq. 7) is trained with a pairwise soft ranking loss over query-image triples and then evaluated against the baselines GNA-RNN, DeViSE, and RelationNet on MyFonts-test. No load-bearing step reduces to a fitted parameter renamed as a prediction, and no uniqueness claim or ansatz is imported from a self-citation. The acknowledged noise and incompleteness of MyFonts web tags is a criterion-validity concern about what is measured, not a circularity in the derivation chain. Author self-citations to DeepFont and related Adobe work are contextual comparisons rather than load-bearing support for the central claim.

Assumptions & free parameters 5 free parameters · 6 assumptions · 2 invented entities

The method introduces two new module types and a large benchmark dataset, but no new physical entities. The central claim rests on the quality of web-sourced font tags, the validity of the rendering process, and the effectiveness of the auxiliary reconstruction and attention losses. The free parameters are hyperparameters chosen by hand; none are estimated from the evaluation test labels.

free parameters (5)
  • lambda (L1 loss weight) = 10
    Weight for the L1 reconstruction loss in Eq. 4, hand-set in Section 5.2.
  • beta (GAN loss weight) = 0.04
    Weight for the GAN loss in Eq. 5, hand-set in Section 5.2.
  • alpha (power activation exponent) = 0.1
    Exponent in the power activation function for the retrieval model, hand-set in Section 5.2 and claimed superior without a comparison.
  • gamma (ranking loss scale) = 100
    Scale in the soft ranking loss of Eq. 7, hand-set in Section 5.2.
  • J (number of glyphs for attention aggregation) = 4
    Number of randomly selected glyph images used to aggregate attention maps during training, Section 4.3.
assumptions (6)
  • domain assumption MyFonts web tags, after preprocessing, provide valid ground truth for font-tag relations.
    Training loss (Eq. 1) and MyFonts-test metrics assume these labels are correct enough. The paper acknowledges noise and applies preprocessing, but the fundamental reliability of the source is assumed.
  • domain assumption Font images are synthetic and controllable by the learning algorithm, enabling the generative reconstruction objective.
    Section 4.2 relies on the ability to render arbitrary characters of the same font and a fixed standard font, which is unique to synthetic image domains.
  • ad hoc to paper The L1 and GAN reconstruction losses force the hidden feature to be font-specific and character-agnostic.
    This is the core assumption of the generative feature learning method. It is motivated by qualitative reconstruction results (Figure 9) but not proven.
  • ad hoc to paper The font class probability distribution is a useful conditioning signal for attention.
    Section 4.3 assumes that the predicted font identity distribution can guide feature re-weighting. The ablation supports it empirically, but the mechanism is not theoretically justified.
  • domain assumption Rendering each font family using its regular version with random character strings represents the font's visual identity.
    Dataset collection in Section 3 selects one regular version per family and renders random strings, assuming this captures the family's visual style for tagging.
  • domain assumption AMT majority agreement (3/3 workers) defines correct ranking for the human evaluation set.
    Section 3 keeps a group only when all three workers agree, assuming unanimous choice equals ground truth and filtering out ambiguous cases.
invented entities (2)
  • Generative feature learning module independent evidence
    purpose: To learn font features that are robust to character identity by reconstructing a different-character glyph from the feature.
    Ablation (Table 1) and qualitative reconstructions (Figure 9) provide in-paper evidence that the module improves retrieval. No external evidence beyond the paper's experiments.
  • Attentive feature selection module independent evidence
    purpose: To re-weight the hidden feature based on predicted font class distribution for better tag prediction.
    Ablation (Table 1) and the attention visualization (Figure 9) support its contribution. Evidence is internal to the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Large-scale Tag-based Font Retrieval with Generative Feature Learning." pith.science (2026). https://pith.science/paper/P2FGERC2

@misc{pith2026190902072,
  author       = {Pith},
  title        = {Pith review of: Large-scale Tag-based Font Retrieval with Generative Feature Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P2FGERC2}},
  note         = {Machine review of arXiv:1909.02072}
}
read the original abstract

Font selection is one of the most important steps in a design workflow. Traditional methods rely on ordered lists which require significant domain knowledge and are often difficult to use even for trained professionals. In this paper, we address the problem of large-scale tag-based font retrieval which aims to bring semantics to the font selection process and enable people without expert knowledge to use fonts effectively. We collect a large-scale font tagging dataset of high-quality professional fonts. The dataset contains nearly 20,000 fonts, 2,000 tags, and hundreds of thousands of font-tag relations. We propose a novel generative feature learning algorithm that leverages the unique characteristics of fonts. The key idea is that font images are synthetic and can therefore be controlled by the learning algorithm. We design an integrated rendering and learning process so that the visual feature from one image can be used to reconstruct another image with different text. The resulting feature captures important font design details while is robust to nuisance factors such as text. We propose a novel attention mechanism to re-weight the visual feature for joint visual-text modeling. We combine the feature and the attention mechanism in a novel recognition-retrieval model. Experimental results show that our method significantly outperforms the state-of-the-art for the important problem of large-scale tag-based font retrieval.

Figures

Figures reproduced from arXiv: 1909.02072 by the authors.

Figure 1
Figure 1. Overview of large-scale tag-based font retrieval. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An example to illustrate that a specific kind of visual fea [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Overview of our large-scale font tagging dataset: (a) A [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Overview of the proposed font retrieval system. We il [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: Combined recognition-retrieval model: the retrieval [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Retrieval results of our model on typical single-tag and multi-tag queries (font is represented by a random 5-character image). [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Comparison of the reconstructed glyph images using [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 11
Figure 11. Figure 11: Qualitative comparison of our basic recognition model [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Group examples of the collected tagging set. For each group of a tag, the ground-truth font is in the red box. [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 13
Figure 13. Figure 13: Font retrieval results of the proposed model on typical single-tag and multi-tag queries. [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 31 canonical work pages

  1. [1]

    Bottom-up and top-down attention for image captioning and vqa

    Peter Anderson, Xiaodong He, Chris Buehler, Damien Teney, Mark Johnson, Stephen Gould, and Lei Zhang. Bottom-up and top-down attention for image captioning and vqa. arXiv preprint arXiv:1707.07998, 2017

  2. [2]

    Multi-content gan for few-shot font style transfer

    Samaneh Azadi, Matthew Fisher, Vladimir Kim, Zhaowen Wang, Eli Shechtman, and Trevor Darrell. Multi-content gan for few-shot font style transfer. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, volume 11, page 13, 2018

  3. [3]

    Large-scale visual font recognition

    Guang Chen, Jianchao Yang, Hailin Jin, Jonathan Brandt, Eli Shechtman, Aseem Agarwala, and Tony X Han. Large-scale visual font recognition. In Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 3598–3605, 2014

  4. [4]

    Greedy function approximation: a gra- dient boosting machine

    Jerome H Friedman. Greedy function approximation: a gra- dient boosting machine. Annals of statistics , pages 1189– 1232, 2001

  5. [5]

    Devise: A deep visual- semantic embedding model

    Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Tomas Mikolov, et al. Devise: A deep visual- semantic embedding model. In Advances in neural informa- tion processing systems, pages 2121–2129, 2013

  6. [6]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pages 2672–2680, 2014

  7. [7]

    Look, imagine and match: Improving textual-visual cross-modal retrieval with generative models

    Jiuxiang Gu, Jianfei Cai, Shafiq Joty, Li Niu, and Gang Wang. Look, imagine and match: Improving textual-visual cross-modal retrieval with generative models. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 7181–7189, 2018

  8. [8]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

Show all 40 references
  1. [9]

    Image-to-image translation with conditional adver- sarial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adver- sarial networks. arXiv preprint, 2017

  2. [10]

    Multimodal residual learning for visual qa

    Jin-Hwa Kim, Sang-Woo Lee, Donghyun Kwak, Min-Oh Heo, Jeonghee Kim, Jung-Woo Ha, and Byoung-Tak Zhang. Multimodal residual learning for visual qa. In Advances in Neural Information Processing Systems , pages 361–369, 2016

  3. [11]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural net- works. In Advances in neural information processing sys- tems, 2012

  4. [12]

    Learning to detect unseen object classes by between- class attribute transfer

    Christoph H Lampert, Hannes Nickisch, and Stefan Harmel- ing. Learning to detect unseen object classes by between- class attribute transfer. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pages 951–958. IEEE, 2009

  5. [13]

    Stacked cross attention for image-text matching

    Kuang-Huei Lee, Xi Chen, Gang Hua, Houdong Hu, and Xi- aodong He. Stacked cross attention for image-text matching. arXiv preprint arXiv:1803.08024, 2018

  6. [14]

    Identity-aware textual-visual matching with latent co-attention

    Shuang Li, Tong Xiao, Hongsheng Li, Wei Yang, and Xi- aogang Wang. Identity-aware textual-visual matching with latent co-attention. In Computer Vision (ICCV), 2017 IEEE International Conference on, pages 1908–1917. IEEE, 2017

  7. [15]

    Person search with natural lan- guage description

    Shuang Li, Tong Xiao, Hongsheng Li, Bolei Zhou, Dayu Yue, and Xiaogang Wang. Person search with natural lan- guage description. arXiv preprint arXiv:1702.05729, 2017

  8. [16]

    Multi-task adversarial network for disentangled feature learning

    Yang Liu, Zhaowen Wang, Hailin Jin, and Ian Was- sell. Multi-task adversarial network for disentangled feature learning. In CVPR, 2018

  9. [17]

    Syn- thetically supervised feature learning for scene text recog- nition

    Yang Liu, Zhaowen Wang, Hailin Jin, and Ian Wassell. Syn- thetically supervised feature learning for scene text recog- nition. In The European Conference on Computer Vision (ECCV), September 2018

  10. [18]

    Knowing when to look: Adaptive attention via a visual sen- tinel for image captioning

    Jiasen Lu, Caiming Xiong, Devi Parikh, and Richard Socher. Knowing when to look: Adaptive attention via a visual sen- tinel for image captioning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 6, page 2, 2017

  11. [19]

    Da-gan: Instance-level image translation by deep attention generative adversarial networks

    Shuang Ma, Jianlong Fu, Chang Wen Chen, and Tao Mei. Da-gan: Instance-level image translation by deep attention generative adversarial networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 5657–5666, 2018

  12. [20]

    Conditional generative adversarial nets

    Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784, 2014

  13. [21]

    Exploratory font selection using crowd- sourced attributes

    Peter O’Donovan, J ¯anis L ¯ıbeks, Aseem Agarwala, and Aaron Hertzmann. Exploratory font selection using crowd- sourced attributes. ACM Transactions on Graphics (TOG) , 33(4):92, 2014

  14. [22]

    Learning deep representations of fine-grained visual descrip- tions

    Scott Reed, Zeynep Akata, Honglak Lee, and Bernt Schiele. Learning deep representations of fine-grained visual descrip- tions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 49–58, 2016

  15. [23]

    Zero-shot learning through cross-modal transfer

    Richard Socher, Milind Ganjoo, Christopher D Manning, and Andrew Ng. Zero-shot learning through cross-modal transfer. In Advances in neural information processing sys- tems, pages 935–943, 2013

  16. [24]

    Disentangled rep- resentation learning gan for pose-invariant face recognition

    Luan Tran, Xi Yin, and Xiaoming Liu. Disentangled rep- resentation learning gan for pose-invariant face recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1415–1424, 2017

  17. [25]

    C. Wah, S. Branson, P. Welinder, P. Perona, and S. Belongie. The Caltech-UCSD Birds-200-2011 Dataset. Technical Re- port CNS-TR-2011-001, California Institute of Technology, 2011

  18. [26]

    Adversarial cross-modal retrieval

    Bokun Wang, Yang Yang, Xing Xu, Alan Hanjalic, and Heng Tao Shen. Adversarial cross-modal retrieval. In Pro- ceedings of the 2017 ACM on Multimedia Conference, pages 154–162. ACM, 2017

  19. [27]

    Deepfont: Identify your font from an image

    Zhangyang Wang, Jianchao Yang, Hailin Jin, Eli Shechtman, Aseem Agarwala, Jonathan Brandt, and Thomas S Huang. Deepfont: Identify your font from an image. In Proceedings of the 23rd ACM international conference on Multimedia , pages 451–459. ACM, 2015

  20. [28]

    Zero- shot learning-the good, the bad and the ugly

    Yongqin Xian, Bernt Schiele, and Zeynep Akata. Zero- shot learning-the good, the bad and the ugly. arXiv preprint arXiv:1703.04394, 2017

  21. [29]

    Learning to compare: Re- lation network for few-shot learning

    Flood Sung Yongxin Yang, Li Zhang, Tao Xiang, Philip HS Torr, and Timothy M Hospedales. Learning to compare: Re- lation network for few-shot learning. 2018

  22. [30]

    Stacked attention networks for image question answering

    Zichao Yang, Xiaodong He, Jianfeng Gao, Li Deng, and Alex Smola. Stacked attention networks for image question answering. In Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition, pages 21–29, 2016

  23. [31]

    Image captioning with semantic attention

    Quanzeng You, Hailin Jin, Zhaowen Wang, Chen Fang, and Jiebo Luo. Image captioning with semantic attention. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4651–4659, 2016

  24. [32]

    Learning a deep embedding model for zero-shot learning

    Li Zhang, Tao Xiang, Shaogang Gong, et al. Learning a deep embedding model for zero-shot learning. 2017

  25. [33]

    Attention-aware deep adversarial hashing for cross-modal retrieval

    Xi Zhang, Hanjiang Lai, and Jiashi Feng. Attention-aware deep adversarial hashing for cross-modal retrieval. In Eu- ropean Conference on Computer Vision , pages 614–629. Springer, 2018

  26. [34]

    Separating style and content for generalized style transfer

    Yexun Zhang, Ya Zhang, and Wenbin Cai. Separating style and content for generalized style transfer. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, volume 1, 2018

  27. [35]

    Dual-path convolutional image-text em- bedding

    Zhedong Zheng, Liang Zheng, Michael Garrett, Yi Yang, and Yi-Dong Shen. Dual-path convolutional image-text em- bedding. arXiv preprint arXiv:1711.05535, 2017

  28. [36]

    Places: A 10 million image database for scene recognition

    Bolei Zhou, Agata Lapedriza, Aditya Khosla, Aude Oliva, and Antonio Torralba. Places: A 10 million image database for scene recognition. IEEE transactions on pattern analysis and machine intelligence, 40(6):1452–1464, 2018

  29. [37]

    Unpaired image-to-image translation using cycle- consistent adversarial networks

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle- consistent adversarial networks. arXiv preprint, 2017

  30. [38]

    Font recognition based on global texture analysis

    Yong Zhu, Tieniu Tan, and Yunhong Wang. Font recognition based on global texture analysis. IEEE Transactions on pat- tern analysis and machine intelligence , 23(10):1192–1200, 2001

  31. [39]

    Optical font recogni- tion using typographical features

    Abdelwahab Zramdini and Rolf Ingold. Optical font recogni- tion using typographical features. IEEE Transactions on Pat- tern Analysis & Machine Intelligence, (8):877–882, 1998

  32. [40]

    Large-scale Tag-based Font Retrieval with Generative Feature Learning

    Supplementary Material In this document, we provide additional materials to sup- plement our paper “Large-scale Tag-based Font Retrieval with Generative Feature Learning”. In the first section, we provide more details about our collected font retrieval dataset. In the second se...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.