pith. machine review for the scientific record. sign in

arxiv: 2605.05865 · v1 · submitted 2026-05-07 · 💻 cs.CV

Recognition: unknown

InkDiffuser: High-Fidelity One-shot Chinese Calligraphy via Differentiable Morphological Optimization

Jing Zhang, Kunchong Shi

Authors on Pith no claims yet

Pith reviewed 2026-05-09 16:27 UTC · model grok-4.3

classification 💻 cs.CV
keywords Chinese calligraphyone-shot generationdiffusion modelsdifferentiable morphologyink renderingfont synthesishigh-frequency enhancement
0
0 comments X

The pith

A diffusion model fuses high-frequency details and uses a differentiable ink loss to generate realistic one-shot Chinese calligraphy.

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

Current methods for generating Chinese calligraphy produce weak stroke rendering and unrealistic ink effects. InkDiffuser addresses this by enhancing content extraction through explicit fusion of high-frequency representations from a single reference glyph and by adding a differentiable ink structure loss that incorporates morphological operations directly into the diffusion process. The loss allows the model to decompose ink traces explicitly, refining stroke contours for greater fluidity and authenticity. A sympathetic reader would care because this reduces the data barrier for high-quality digital calligraphy while preserving artistic visual properties.

Core claim

By fusing high-frequency representations to capture accurate font structure and introducing a Differentiable Ink Structure loss that integrates differentiable morphological operations into diffusion, InkDiffuser decomposes ink-trace structures explicitly and refines stroke contours, enabling generation of calligraphy with realistic ink morphology, structural consistency, and visual authenticity from only a single reference glyph.

What carries the argument

The Differentiable Ink Structure (DIS) loss, which embeds differentiable morphological operations inside the diffusion training loop to enforce explicit decomposition of ink morphology for contour-level refinement.

If this is right

  • One-shot synthesis produces calligraphy with stroke and ink quality that exceeds prior few-shot font generators.
  • Explicit morphological regularization inside diffusion improves fine detail without separate post-processing steps.
  • Complex characters maintain structural integrity and artistic fluidity under the same single-glyph training regime.
  • The framework demonstrates that ink-trace decomposition can be learned end-to-end rather than through separate rendering modules.

Where Pith is reading between the lines

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

  • The same morphological loss pattern could be tested on other stroke-based arts such as Japanese kanji or brush painting to check transfer of realism gains.
  • If the high-frequency fusion step proves stable, it may reduce the need for large multi-style datasets in related generative design tasks.
  • Deployment in digital art tools could allow users to iterate on calligraphy styles with far fewer reference images than current pipelines require.

Load-bearing premise

Explicitly adding high-frequency fusion and differentiable morphological operations to a diffusion model will raise ink realism and stroke fidelity without introducing new artifacts or demanding heavy per-style adjustments.

What would settle it

Side-by-side expert ratings or pixel-level ink morphology measurements on complex characters where InkDiffuser outputs receive lower authenticity scores or display more contour artifacts than the strongest existing few-shot baselines.

Figures

Figures reproduced from arXiv: 2605.05865 by Jing Zhang, Kunchong Shi.

Figure 1
Figure 1. Figure 1: Comparison between real calligraphy (bottom) and results of existing view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed InkDiffuser framework. Given a content image view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the sigmoid function σ(z/τ) with varying temperature parameters τ. As τ decreases, the curve approaches a sharp step function, allowing for a differentiable approximation of the hard logic operators. In our DIS Loss, we implement these soft morphological operations. The convolution operation, conv(x, kernel), ag￾gregates information from the neighborhood defined by the circular structuring … view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on the UFUC setting. The results generated by InkDiffuser are closest to the target fonts and contain noticeably fewer wrongly view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results of the ablation study on the STAF module. view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results of the ablation study on the DIS module. view at source ↗
read the original abstract

Current Chinese calligraphy generation methods suffer from poor stroke rendering and unrealistic ink morphology, resulting in outputs with limited visual fidelity and artistic fluidity. To address this problem, we propose \textbf{InkDiffuser}, a diffusion-based generative framework for one-shot Chinese calligraphy synthesis. To guarantee high-fidelity rendering, we introduce two core contributions: a high-frequency enhancement mechanism and a Differentiable Ink Structure (DIS) loss that explicitly regularizes ink morphology. Inspired by the observation that high-frequency information in individual samples typically carries contour details, we enhance content extraction by explicitly fusing high-frequency representations for more accurate font structure. Furthermore, we propose a differentiable ink structure loss that integrates differentiable morphological operations into the diffusion process. By allowing the model to learn an explicit decomposition of ink-trace structures, DIS facilitates fine-grained refinement of stroke contours and delivers significantly improved visual realism in the generated calligraphy. Extensive experiments on various calligraphic styles and complex characters demonstrate that InkDiffuser can generate superior calligraphy fonts with realistic ink rendering effects from only a single reference glyph and outperform existing few-shot font generation approaches in structural consistency, detail fidelity, and visual authenticity. The code is available at the following address: https://github.com/JingVIPLab/InkDiffuser.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces InkDiffuser, a diffusion-based framework for one-shot Chinese calligraphy generation. Key contributions include a high-frequency enhancement mechanism for better content extraction from reference glyphs and a Differentiable Ink Structure (DIS) loss that integrates differentiable morphological operations to regularize ink morphology and refine stroke contours. The authors claim that this approach generates superior calligraphy with realistic ink effects from a single reference and outperforms existing few-shot font generation methods in structural consistency, detail fidelity, and visual authenticity, as demonstrated through extensive experiments on various calligraphic styles and complex characters.

Significance. Should the results hold, this could advance the field of generative AI for traditional arts by addressing specific challenges in ink rendering and stroke fidelity in calligraphy synthesis. The differentiable morphological optimization in diffusion models offers a novel regularization strategy that may inspire similar techniques in other domains requiring fine structural control. The public code release supports reproducibility and further research.

major comments (2)
  1. [DIS loss (methods section)] The Differentiable Ink Structure (DIS) loss is presented as integrating differentiable morphological operations into the diffusion process to enable explicit decomposition of ink-trace structures for fine-grained stroke contour refinement. However, standard morphological operations (erosion/dilation) are non-differentiable, and any implementation requires relaxations such as soft min/max or neural approximations. The manuscript must explicitly describe the chosen relaxation method (likely in the methods section defining the DIS loss) and provide evidence—such as gradient analysis, ablation studies, or visualization of morphology adjustments—that these gradients meaningfully refine ink morphology on real calligraphy traces without introducing new artifacts. If the relaxations fail to deliver effective regularization, the one-shot superiority claim over standard diffusion objectives would
  2. [Experimental evaluation (results section)] The abstract and summary assert that 'extensive experiments' demonstrate outperformance in structural consistency, detail fidelity, and visual authenticity over existing few-shot approaches. No quantitative metrics, baseline comparisons, tables, FID scores, user-study results, or error analysis are referenced. This absence is load-bearing for the central claim of superiority, as visual claims in calligraphy generation are inherently subjective without supporting data or protocols.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the DIS loss formulation and the need for quantitative evaluation. We address each major comment below and will revise the manuscript to strengthen the presentation of our contributions.

read point-by-point responses

  1. Referee: [DIS loss (methods section)] The Differentiable Ink Structure (DIS) loss is presented as integrating differentiable morphological operations into the diffusion process to enable explicit decomposition of ink-trace structures for fine-grained stroke contour refinement. However, standard morphological operations (erosion/dilation) are non-differentiable, and any implementation requires relaxations such as soft min/max or neural approximations. The manuscript must explicitly describe the chosen relaxation method (likely in the methods section defining the DIS loss) and provide evidence—such as gradient analysis, ablation studies, or visualization of morphology adjustments—that these gradients meaningfully refine ink morphology on real calligraphy traces without introducing new artifacts. If the relaxations fail to deliver effective regularization, the one-shot superiority claim over

    Authors: We agree that the relaxation method for differentiability must be described explicitly. The DIS loss in the manuscript employs soft morphological operations using differentiable approximations to min/max via a temperature-controlled softmin function (similar to standard relaxations in differentiable morphology literature). However, we acknowledge the current description in the methods section is insufficiently detailed. In the revised manuscript, we will expand the DIS loss definition with the full mathematical formulation of the soft erosion/dilation, include an ablation study isolating the effect of DIS on stroke refinement, and add visualizations of gradient magnitudes and morphology adjustments on real calligraphy traces to confirm effective regularization without new artifacts. This will directly support the one-shot superiority claims. revision: yes

  2. Referee: [Experimental evaluation (results section)] The abstract and summary assert that 'extensive experiments' demonstrate outperformance in structural consistency, detail fidelity, and visual authenticity over existing few-shot approaches. No quantitative metrics, baseline comparisons, tables, FID scores, user-study results, or error analysis are referenced. This absence is load-bearing for the central claim of superiority, as visual claims in calligraphy generation are inherently subjective without supporting data or protocols.

    Authors: We agree that the absence of quantitative metrics weakens the superiority claims, as visual inspection alone is subjective. Although the manuscript presents extensive qualitative comparisons across calligraphic styles and complex characters, it lacks numerical tables and protocols. In the revised version, we will add a dedicated quantitative evaluation subsection with FID, LPIPS, and SSIM scores against few-shot baselines, plus a user study protocol and results measuring structural consistency, detail fidelity, and perceived authenticity. This will provide objective support for the claims made in the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper proposes InkDiffuser as a new diffusion framework whose two core technical contributions—a high-frequency fusion mechanism for contour extraction and the Differentiable Ink Structure (DIS) loss that inserts differentiable morphological operations—are introduced as explicit, novel regularizers rather than as redefinitions or fits of the target quantity. No equation or claim reduces the generated calligraphy output to a parameter fitted on the same data, nor does any load-bearing step rest on a self-citation whose content is itself unverified within the paper. The asserted improvements in stroke fidelity and ink realism are presented as empirical consequences of these added mechanisms, not as identities that hold by construction. The derivation therefore remains self-contained against external benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Without the full manuscript, specific free parameters, axioms, or invented entities cannot be extracted; the abstract introduces a new loss term but does not detail any fitted constants or background assumptions.

pith-pipeline@v0.9.0 · 5516 in / 1118 out tokens · 52568 ms · 2026-05-09T16:27:10.768727+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 6 canonical work pages · 2 internal anchors

  1. [1]

    Few-shot font generation by learning fine-grained local styles,

    L. Tang, Y . Cai, J. Liu, Z. Hong, M. Gong, M. Fan, J. Han, J. Liu, E. Ding, and J. Wang, “Few-shot font generation by learning fine-grained local styles,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 7895–7904

    2022

  2. [2]

    Few-shot com- positional font generation with dual memory,

    J. Cha, S. Chun, G. Lee, B. Lee, S. Kim, and H. Lee, “Few-shot com- positional font generation with dual memory,” inEuropean conference on computer vision. Springer, 2020, pp. 735–751

    2020

  3. [3]

    Deepcallifont: Few-shot chinese calligraphy font synthesis by integrating dual-modality generative models,

    Y . Liu and Z. Lian, “Deepcallifont: Few-shot chinese calligraphy font synthesis by integrating dual-modality generative models,” inProceed- ings of the AAAI conference on artificial intelligence, vol. 38, no. 4, 2024, pp. 3774–3782

    2024

  4. [4]

    Scfont: Structure-guided chinese font generation via deep stacked networks,

    Y . Jiang, Z. Lian, Y . Tang, and J. Xiao, “Scfont: Structure-guided chinese font generation via deep stacked networks,” inProceedings of the AAAI conference on artificial intelligence, vol. 33, no. 01, 2019, pp. 4015– 4022

    2019

  5. [5]

    Tet-gan: Text effects transfer via stylization and destylization,

    S. Yang, J. Liu, W. Wang, and Z. Guo, “Tet-gan: Text effects transfer via stylization and destylization,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 1238–1245

    2019

  6. [6]

    Calliffusion: Chinese calligraphy generation and style transfer with diffusion modeling,

    Q. Liao, G. Xia, and Z. Wang, “Calliffusion: Chinese calligraphy generation and style transfer with diffusion modeling,”arXiv preprint arXiv:2305.19124, 2023

    work page arXiv 2023

  7. [7]

    Handwritten chinese font generation with collaborative stroke refinement,

    C. Wen, Y . Pan, J. Chang, Y . Zhang, S. Chen, Y . Wang, M. Han, and Q. Tian, “Handwritten chinese font generation with collaborative stroke refinement,” inProceedings of the IEEE/CVF winter conference on applications of computer vision, 2021, pp. 3882–3891

    2021

  8. [8]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems, vol. 33, pp. 6840– 6851, 2020

    2020

  9. [9]

    Denoising Diffusion Implicit Models

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” arXiv preprint arXiv:2010.02502, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  10. [10]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differ- ential equations,”arXiv preprint arXiv:2011.13456, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  11. [11]

    Diff- font: Diffusion model for robust one-shot font generation,

    H. He, X. Chen, C. Wang, J. Liu, B. Du, D. Tao, and Q. Yu, “Diff- font: Diffusion model for robust one-shot font generation,”International Journal of Computer Vision, vol. 132, no. 11, pp. 5372–5386, 2024

    2024

  12. [12]

    Mx-font++: Mixture of heterogeneous aggregation experts for few-shot font generation,

    W. Wang, D. Sun, J. Zhang, and L. Gao, “Mx-font++: Mixture of heterogeneous aggregation experts for few-shot font generation,” in ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025, pp. 1–5

    2025

  13. [13]

    Fontdiffuser: One-shot font generation via denoising diffusion with multi-scale content aggregation and style contrastive learning,

    Z. Yang, D. Peng, Y . Kong, Y . Zhang, C. Yao, and L. Jin, “Fontdiffuser: One-shot font generation via denoising diffusion with multi-scale content aggregation and style contrastive learning,” inProceedings of the AAAI conference on artificial intelligence, vol. 38, no. 7, 2024, pp. 6603–6611

    2024

  14. [14]

    Few shot font gen- eration via transferring similarity guided global style and quantization local style,

    W. Pan, A. Zhu, X. Zhou, B. K. Iwana, and S. Li, “Few shot font gen- eration via transferring similarity guided global style and quantization local style,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 19 506–19 516

    2023

  15. [15]

    Few-shot font generation by learning style difference and similarity,

    X. He, M. Zhu, N. Wang, and X. Gao, “Few-shot font generation by learning style difference and similarity,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 9, pp. 8013–8025, 2024

    2024

  16. [16]

    Hgan: Hierarchical graph alignment network for image-text retrieval,

    J. Guo, M. Wang, Y . Zhou, B. Song, Y . Chi, W. Fan, and J. Chang, “Hgan: Hierarchical graph alignment network for image-text retrieval,” IEEE Transactions on Multimedia, vol. 25, pp. 9189–9202, 2023

    2023

  17. [17]

    Calligan: Style and structure-aware chinese calligraphy character generator,

    S.-J. Wu, C.-Y . Yang, and J. Y .-j. Hsu, “Calligan: Style and structure-aware chinese calligraphy character generator,”arXiv preprint arXiv:2005.12500, 2020

    work page arXiv 2005

  18. [18]

    Stargan: Unified generative adversarial networks for multi-domain image-to- image translation,

    Y . Choi, M. Choi, M. Kim, J.-W. Ha, S. Kim, and J. Choo, “Stargan: Unified generative adversarial networks for multi-domain image-to- image translation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 8789–8797

    2018

  19. [19]

    Gan electronics,

    S. J. Pearton and F. Ren, “Gan electronics,”Advanced Materials, vol. 12, no. 21, pp. 1571–1580, 2000

    2000

  20. [20]

    Image-to-image translation with conditional adversarial networks,

    P. Isola, J.-Y . Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1125– 1134

    2017

  21. [21]

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

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 2223–2232

    2017

  22. [22]

    Learning to write stylized chinese characters by reading a handful of examples,

    D. Sun, T. Ren, C. Li, H. Su, and J. Zhu, “Learning to write stylized chinese characters by reading a handful of examples,”arXiv preprint arXiv:1712.06424, 2017

    work page arXiv 2017

  23. [23]

    Separating style and content for generalized style transfer,

    Y . Zhang, Y . Zhang, and W. Cai, “Separating style and content for generalized style transfer,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 8447–8455

    2018

  24. [24]

    Artistic glyph image syn- thesis via one-stage few-shot learning,

    Y . Gao, Y . Guo, Z. Lian, Y . Tang, and J. Xiao, “Artistic glyph image syn- thesis via one-stage few-shot learning,”ACM Transactions on Graphics (ToG), vol. 38, no. 6, pp. 1–12, 2019

    2019

  25. [25]

    Dg-font: Deformable generative networks for unsupervised font generation,

    Y . Xie, X. Chen, L. Sun, and Y . Lu, “Dg-font: Deformable generative networks for unsupervised font generation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 5130–5140

    2021

  26. [26]

    Few-shot font generation with localized style representations and factorization,

    S. Park, S. Chun, J. Cha, B. Lee, and H. Shim, “Few-shot font generation with localized style representations and factorization,” inProceedings of the AAAI conference on artificial intelligence, vol. 35, no. 3, 2021, pp. 2393–2402

    2021

  27. [27]

    Xmp-font: Self-supervised cross-modality pre-training for few-shot font generation,

    W. Liu, F. Liu, F. Ding, Q. He, and Z. Yi, “Xmp-font: Self-supervised cross-modality pre-training for few-shot font generation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 7905–7914

    2022

  28. [28]

    Cf-font: Content fusion for few-shot font generation,

    C. Wang, M. Zhou, T. Ge, Y . Jiang, H. Bao, and W. Xu, “Cf-font: Content fusion for few-shot font generation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 1858–1867

    2023

  29. [29]

    If-font: Ideographic description sequence- following font generation,

    X. Chen, X. Ke, and W. Guo, “If-font: Ideographic description sequence- following font generation,”Advances in Neural Information Processing Systems, vol. 37, pp. 14 177–14 199, 2024

    2024

  30. [30]

    Vq-font: Few-shot font generation with structure-aware enhancement and quantization,

    M. Yao, Y . Zhang, X. Lin, X. Li, and W. Zuo, “Vq-font: Few-shot font generation with structure-aware enhancement and quantization,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 15, 2024, pp. 16 407–16 415

    2024

  31. [31]

    Few-shot unsupervised image-to-image translation,

    M.-Y . Liu, X. Huang, A. Mallya, T. Karras, T. Aila, J. Lehtinen, and J. Kautz, “Few-shot unsupervised image-to-image translation,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 10 551–10 560

    2019

  32. [32]

    Multi-content gan for few-shot font style transfer,

    S. Azadi, M. Fisher, V . G. Kim, Z. Wang, E. Shechtman, and T. Darrell, “Multi-content gan for few-shot font style transfer,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7564–7573

    2018

  33. [33]

    Dp- font: Chinese calligraphy font generation using diffusion model and physical information neural network,

    L. Zhang, Y . Zhu, A. Benarab, Y . Ma, Y . Dong, and J. Sun, “Dp- font: Chinese calligraphy font generation using diffusion model and physical information neural network,” inProceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, IJCAI-24, K. Larson, Ed. International Joint Conferences on Artificial Intelligence Organiza...

    2024

  34. [34]

    Unit-ddpm: Unpaired image translation with denoising diffusion probabilistic models,

    H. Sasaki, C. G. Willcocks, and T. P. Breckon, “Unit-ddpm: Unpaired image translation with denoising diffusion probabilistic models,”arXiv preprint arXiv:2104.05358, 2021

    work page arXiv 2021

  35. [35]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inInternational Conference on Medical image computing and computer-assisted intervention. Springer, 2015, pp. 234–241

    2015

  36. [36]

    On the spectral bias of neural networks,

    N. Rahaman, A. Baratin, D. Arpit, F. Draxler, M. Lin, F. Hamprecht, Y . Bengio, and A. Courville, “On the spectral bias of neural networks,” inInternational conference on machine learning. PMLR, 2019, pp. 5301–5310

    2019

  37. [37]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 3–19

    2018

  38. [38]

    The coefficient of determi- nation r-squared is more informative than smape, mae, mape, mse and rmse in regression analysis evaluation,

    D. Chicco, M. J. Warrens, and G. Jurman, “The coefficient of determi- nation r-squared is more informative than smape, mae, mape, mse and rmse in regression analysis evaluation,”Peerj computer science, vol. 7, p. e623, 2021

    2021

  39. [39]

    Efficient dilation, erosion, opening, and clos- ing algorithms,

    J. Y . Gil and R. Kimmel, “Efficient dilation, erosion, opening, and clos- ing algorithms,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 12, pp. 1606–1617, 2003. 10

    2003

  40. [40]

    The influence of the sigmoid function pa- rameters on the speed of backpropagation learning,

    J. Han and C. Moraga, “The influence of the sigmoid function pa- rameters on the speed of backpropagation learning,” inInternational workshop on artificial neural networks. Springer, 1995, pp. 195–201

    1995

  41. [41]

    Laplacian matrices of graphs: a survey,

    R. Merris, “Laplacian matrices of graphs: a survey,”Linear algebra and its applications, vol. 197, pp. 143–176, 1994

    1994

  42. [42]

    Foundertype font library,

    Beijing Founder Electronics Co., Ltd., “Foundertype font library,” https: //www.foundertype.com/, 2024, accessed: 2025-12-20

    2024

  43. [43]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  44. [44]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    2018

  45. [45]

    Generate like experts: Multi-stage font generation by incorporating font transfer pro- cess into diffusion models,

    B. Fu, F. Yu, A. Liu, Z. Wang, J. Wen, J. He, and Y . Qiao, “Generate like experts: Multi-stage font generation by incorporating font transfer pro- cess into diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 6892–6901

    2024