Pith. sign in

REVIEW 3 major objections 6 minor 64 references

Design-o-meter: Towards Evaluating and Refining Graphic Designs

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Design-o-meter claims that one unified, roughly 410k-parameter framework can both score graphic-design quality and refine layouts to improve its own score, beating multimodal LLMs like GPT-4o and prior refinement systems on Crello…

desk verdict A plausible scorer and a clever refiner, but the refinement loop is closed: no human evaluation, and the 'first unified framework' claim ignores Aesthetics++. read the letter →

arxiv 2411.14959 v1 pith:OW4LHLII submitted 2024-11-22 cs.CV cs.AIcs.HC

Sahil Goyal , Abhinav Mahajan , Swasti Mishra , Prateksha Udhayanan , Tripti Shukla , K J Joseph , Balaji Vasan Srinivasan This is my paper · ORCID
classification cs.CVcs.AIcs.HC
keywords designscoringrefinementgraphicSiamesenetworkmetriclearninggeneticalgorithmlayoutaestheticsmultimodalLLMevaluation
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

Design-o-meter tries to establish that graphic-design quality can be measured and improved by one unified, data-driven framework, without human aesthetic ratings. The paper argues that a small contrastively trained scorer captures layout nuances that recent multimodal LLMs miss, and that a genetic algorithm using that scorer as its fitness function can move elements to raise the score. If correct, this gives generative-design pipelines an automatic, reference-free quality signal and a practical editing step, replacing expensive LLM evaluation and slow heuristic optimization.

What carries the argument

The machinery is the pair of a scorer and a crossover operator. The scorer is a four-layer convolutional network with group normalization, totaling about 410k parameters, that takes the rendered image concatenated with a color-coded layout map and learns to rank good designs above bad ones rather than predict human scores. SWAN is the design-specific crossover that replaces blind copy-paste in the genetic algorithm: it builds grid lines from existing element edges, finds the most size- and proximity-similar box for an incoming image or SVG, snaps it in, aligns text elements by minimum Euler distance to the nearest element, and only mutates within canvas bounds. The layout encoding lets the scorer attend to overlaps and alignments, and SWAN is what lets the genetic search navigate the design space efficiently.

What would settle it

A human preference study: if annotators shown (original, perturbed) and (original, refined) pairs do not systematically prefer the originals over the perturbed versions, or prefer the original over Design-o-meter's refinement, then the perturbation recipe does not track human aesthetic judgment and the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that scoring and refining can share one loop: a design is represented as structured metadata, rendered to an image and paired with a color-coded layout map that marks text, image, and overlap regions. The scorer is a Siamese metric-learning network trained contrastively on Crello templates (filtered to at most ten elements and no text-image overlap) as good, and on 22 location and scale perturbations as bad. The refiner is NSGA-II with a design-specific crossover, SWAN, that snaps image elements into grid slots and aligns text elements to nearby elements, with the scorer as the single objective. Reported results include rank accuracy of 94.97 on biased pairs and 87.50 on unbiased cross-match pairs versus 68.84 and 63.50 for GPT-4o, and refinement T-mIoU of 54.44 in the refine-all setting against a previous best of 50.08.

Load-bearing premise

The load-bearing premise is that a design is bad exactly when one of the 22 position- or scale-perturbations has been applied to an otherwise acceptable Crello template, so the scorer learns that synthetic dichotomy and the refiner optimizes it.

Editorial extensions

If this is right

  • Design quality can be scored without human annotations or reference designs, so generated layouts can be filtered at production speed.
  • Refinement becomes closed-loop: any design can be edited by optimizing the scorer, and the same score that judges the result also guides the search.
  • A model with about 410k parameters can outperform billion-parameter LLM evaluators at ranking design pairs, making cheap on-device design evaluation feasible.
  • The scorer can serve as an off-the-shelf discriminator or reward model for other layout and design generation frameworks.
  • The refiner's roughly 30-second turnaround brings automated design refinement into interactive editing tools.

Reading between the lines

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

  • If the contrastive recipe transfers, the same pipeline could score and refine color schemes, typography, or content attributes by adding perturbations for those dimensions, since the paper's ablations suggest the scorer already picks up non-layout design principles.
  • The definition of 'bad' is the paper's assumption; a human-preference validation would tell whether the optimized layouts are judged better by people, not just by the scorer.
  • The SWAN and genetic-search pattern is generic: any cheap judge plus grid-snapping crossover could refine other structured outputs such as UI layouts, documents, or slide decks.
  • The reported score jumps from roughly 0.1 to 0.9 in qualitative samples suggest the scorer's scale is not calibrated to human ratings, so using it as a reward could overfit to its own notion of goodness.
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

3 major / 6 minor

Summary. The paper proposes Design-o-meter, a framework consisting of (1) a Siamese-network scorer that ranks graphic designs by 'goodness' and (2) a genetic-algorithm refiner (SWAN) that adjusts layout attributes to maximize the scorer's output. The scorer is trained on pairs of good/bad designs created by filtering Crello templates and applying 22 position/scale perturbations; the refiner is evaluated by reconstructing the original templates from randomly initialized layouts. Experiments compare the scorer with GPT-4o and LLaVA-NeXT on biased, color, and cross-match settings, and the refiner with prior layout-refinement methods on text-box and full-layout tasks. Additional ablations and sensitivity analyses are provided.

Significance. If validated, the framework would be practically useful for automated design feedback and refinement. The authors should be credited for a lightweight (410K-parameter) scorer that outperforms billion-parameter multimodal LLMs on the constructed ranking tasks, for including held-out perturbation settings that show some generalization, and for reporting failure cases and hyperparameter sensitivity. However, the absence of any human evaluation, combined with the synthetic definition of 'badness' and the closed-loop refinement evaluation, means the central claim of measuring and improving aesthetic quality is not yet established. The work as presented is more accurately described as a method for detecting and inverting a specific class of layout perturbations on the Crello dataset.

major comments (3)
  1. [Sec. 3.1, Sec. 4.1 (Table 2)] The definition of 'good' and 'bad' designs is entirely synthetic: good designs are Crello templates passing two arbitrary filters (at most 10 elements, no text-image overlap), and bad designs are those templates after any of 22 position/scale perturbations, or, in the color setting, recoloring to a CIELAB distance of 2-3, which is asserted to be 'unaesthetic' without support. The claim that these perturbations are 'modeled on how humans consider a design good and bad' (Sec. 3.1) is not supported by any citation or experiment. The 'unbiased' color and cross-match evaluations still consist of synthetic pairs derived from the same template pool; they test generalization to other perturbation recipes, not alignment with human aesthetic judgments. Therefore the high RAcc values (94.97, 90.45, 87.50) do not establish that the scorer 'quantifies the goodness' of designs in a human-meaningful sense, which is the premise of the paper's title and abstract.
  2. [Sec. 3.2, Sec. 4.2 (Tables 3 and 4)] The refinement evaluation is a closed loop. The refiner's fitness function is the scorer, and the scorer is trained to prefer original layouts over the exact position/scale perturbations used to create the ground-truth layouts for refinement (mIoU, T-mIoU). Thus the reported refinement successes (e.g., T-mIoU 54.44 in Table 3) are consistent with the system learning to invert the training-time perturbation distribution, rather than improving aesthetic appeal. The score increases shown in Figs. 7 and 8 are self-assessments by the same scorer, not independent evidence. To support the claim that SWAN 'improves visual appeal,' a human preference study comparing original, noisy, and refined designs (or an external aesthetic metric not derived from the training perturbation scheme) is required.
  3. [Sec. 3.1, Sec. 2.1] The paper criticizes Tabata et al. for using random layout movements to create negative examples, noting that such designs 'are not necessarily bad' and 'can sometimes exhibit creative and unique layouts.' The same concern applies to the paper's own perturbation recipe: for instance, moving the largest element or cluttering elements toward a corner may occasionally produce a reasonable layout, and the choice of these 22 transformations as the ground-truth notion of 'badness' is arbitrary. Since the paper's stated goal (Sec. 1) is an evaluation that 'corroborates with human perception,' the absence of any empirical validation that these perturbations are perceived as bad (and that the filtered originals are perceived as good) is a load-bearing gap.
minor comments (6)
  1. [Sec. 3.1, Eq. (3)] The formula has a mismatched parenthesis: it reads `max(0, m− (S(Dgood_meta) − S(Dbad_meta))` with an extra opening parenthesis; it should be `max(0, m − (S(Dgood_meta) − S(Dbad_meta)))` or equivalently `max(0, m − S(Dgood_meta) + S(Dbad_meta))`.
  2. [Sec. 3.1, Eqs. (4)-(5)] The notation `Psim(S(Dgood_meta), S(Dbad_meta))` is misleading because `Psim` is defined in Eq. (5) as a dot product of feature embeddings `F(Dgood_meta)` and `F(Dbad_meta)`, not of the scalar scores; please clarify the notation.
  3. [Sec. 4] The term 'unbiased' is used to mean 'not using the same perturbation transformations as training,' but the data are still synthetic and derived from the same template pool; consider renaming to something like 'held-out perturbation settings' to avoid implying independence from human judgment.
  4. [Abstract, Introduction] The claim of being 'the first approach that scores and refines designs in a unified framework' should be qualified, since Aesthetics++ [30] also combines aesthetic scoring with a refinement mechanism; please clarify the specific sense (e.g., self-supervised, no human annotations, GA-based) in which the novelty holds.
  5. [Throughout] The abbreviation 'SWAN' is sometimes typeset with a space (e.g., 'SW AN' in Sec. 3.2 and the Fig. 2 caption); please unify the notation.
  6. [Table 2] The 'Params' column is blank for GPT-4o; either report the parameter count or explicitly state that it is not publicly available.

Circularity Check

2 steps flagged · score 6.0 of 10

Refinement loop is self-referential: 'good' is defined as unperturbed Crello layouts, the refiner optimizes the scorer trained on that definition, and refinement is evaluated by how well it returns to those same layouts.

  1. self definitional [Sec. 3.1 (Dataset Creation); Sec. 3.2 (Design Refiner); Sec. 4.2 (Refine-All evaluation and Eq. 7)]
    "Given a set of designs from any design dataset [59], we first filter them on two criteria: ... to create a list of good designs Dgood_meta. Next, we surgically modify these designs to make them bad by altering the location and scale of its constituent elements. We employ 22 such types of transformations to create Dbad_meta. ... Our scorer (Sec. 3.1), guides SWAN, to refine its design aesthetics by acting as the objective function being optimized."

    'Good' is defined as an unperturbed Crello template passing two filters; 'bad' is the same template after 22 position/scale perturbations. The scorer is trained on exactly this constructed dichotomy. The refiner then maximizes that same learned scorer, and the refine-all evaluation (T-mIoU, Eq. 7) measures agreement with the same original Crello templates that generated Dgood. Thus the reported 'improvement in visual appeal' operationally reduces to moving a layout back toward the training anchor; the aesthetic claim is not independently anchored. The unbiased color and cross-match tests provide partial generalization evidence, which prevents this from being a full tautology.

  2. fitted input called prediction [Sec. 4.2 (Refine-Text/Refine-All experimental setup and Eq. 7)]
    "We randomly initialize the values of the center coordinates and scale of all elements except the background element. Tab. 3 showcases comparison with 7 baseline approaches ... Following CanvasVAE [59], we use Type-wise mIoU (T-mIoU) to evaluate our refinement module for refine-all settings."

    The refinement task is to reconstruct the original Crello coordinates and scales from random initialization, and the success metric is overlap with those same original layouts. Since the optimizer's fitness is the scorer fitted on original-versus-perturbed pairs, the reported T-mIoU/mIoU numbers measure how well the scorer's landscape points back to the training-good originals. This is a denoising/reconstruction benchmark, not an independent measure of aesthetic improvement; presenting it as 'refinement improves visual appeal' renames the fitted objective as an external quality gain. The baseline comparisons remain meaningful for the reconstruction task itself, but not for the broader aesthetic claim.

full rationale

The core loop is internally closed but not completely tautological. Section 3.1 defines 'good' as filtered Crello templates and 'bad' as the same templates after 22 position/scale perturbations, and trains the scorer on that dichotomy. Section 3.2 makes this scorer the fitness function of the genetic refiner, and Section 4.2 evaluates refinement by IoU/T-mIoU against the same original templates. Consequently, the 'improvement in visual appeal' claim rests on the unvalidated identification of the perturbation recipe with human aesthetics; the qualitative score gains in Figs. 7 and 8 are the optimizer's self-assessment. This is a partial circularity, because a design is 'good' by construction insofar as it lies near the training anchor. The unbiased color and cross-match evaluations (Table 2) do test generalization to different synthetic perturbations and show the scorer is not a lookup table; that independent signal prevents a higher score. No load-bearing self-citation or imported uniqueness theorem was found; references to FlexDM and CanvasVAE are external baselines. The main gap is the absence of any human-subject validation connecting the perturbation-based label to perceived quality, which is exactly the step needed to break the loop.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the perturbation-based definition of good versus bad design, the assumption that position and scale are the actionable aesthetic axes, and a filtering step that limits the scope to simple layouts. These are all introduced by the authors rather than established externally.

free parameters (6)
  • hinge margin m = 0.2
    Hard margin in Eq. 3, tuned on validation; low value achieves highest RAcc 94.97.
  • loss weights alpha, beta = 0.8, 0.2
    Sensitivity analysis in Table 6; decreasing alpha and increasing beta has positive effect.
  • CIELAB distance range for bad colors = (2, 3)
    Used to create the unbiased color dataset; selection of this range is a design choice that may influence RAcc.
  • perturbation noise stds = 0.05, 0.1, 0.2, 0.5
    Varying degrees of badness in location perturbations; these are chosen without justification.
  • GA population size, n_trials, p = 100, 1500, 0.3
    GA hyperparameters selected for good performance; no grid evidence.
  • margin type H-Margin = H-Margin
    Selected among H/TB/Ada margins; H-Margin with 0.2 gave best validation RAcc.
assumptions (4)
  • domain assumption Original Crello designs are 'good' designs; any perturbation of location/scale yields a 'bad' design.
    This defines the ground truth for all scoring and refinement experiments (Sec. 3.1).
  • domain assumption Aesthetic quality is primarily determined by element position and scale; the scorer's layout encoding captures this.
    The refiner only optimizes position/scale attributes, and the scorer is trained on such perturbations (Sec. 3.1, 3.2).
  • ad hoc to paper Max 10 elements and no text-image overlap filter yields representative 'good designs'.
    This selection criterion for good designs may bias the model away from complex real-world layouts (Sec. 3.1).
  • domain assumption GroupNorm is the correct normalization for designs because designs are not from a common distribution.
    Empirical choice; no theoretical justification beyond observed RAcc (Sec. 5.3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Design-o-meter: Towards Evaluating and Refining Graphic Designs." pith.science (2026). https://pith.science/paper/OW4LHLII

@misc{pith2026241114959,
  author       = {Pith},
  title        = {Pith review of: Design-o-meter: Towards Evaluating and Refining Graphic Designs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OW4LHLII}},
  note         = {Machine review of arXiv:2411.14959}
}
read the original abstract

Graphic designs are an effective medium for visual communication. They range from greeting cards to corporate flyers and beyond. Off-late, machine learning techniques are able to generate such designs, which accelerates the rate of content production. An automated way of evaluating their quality becomes critical. Towards this end, we introduce Design-o-meter, a data-driven methodology to quantify the goodness of graphic designs. Further, our approach can suggest modifications to these designs to improve its visual appeal. To the best of our knowledge, Design-o-meter is the first approach that scores and refines designs in a unified framework despite the inherent subjectivity and ambiguity of the setting. Our exhaustive quantitative and qualitative analysis of our approach against baselines adapted for the task (including recent Multimodal LLM-based approaches) brings out the efficacy of our methodology. We hope our work will usher more interest in this important and pragmatic problem setting.

Figures

Figures reproduced from arXiv: 2411.14959 by the authors.

Figure 1
Figure 1. The figure presents an overview of Design-o-meter. It takes a design as input. The scorer evaluates the design and provides a unified design score. The refiner refines the design with the help of the design score to improve its aesthetic appeal. tive AI technologies can work hand-in-hand with designers to supplement them in all phases of their creative workflow. Off-late, such generative models [7, 8, 20, 22, 57] ha… view at source ↗
Figure 2
Figure 2. A visual illustration of SWAN: Design Specific Crossover with Smart Snapping. Given two parents, SWAN first randomly decides which element to pick from either of the parents to generate the child. Then, it copies over the content from the first parent to the current canvas. Next, it identifies potential area within the canvas to host elements from the second parent, guided by grid-lines. Finally, the content from th… view at source ↗
Figure 3
Figure 3. Samples from Crello dataset and their layout encodings. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Qualitative results of Design-o-meter refining graphic designs. The top sub-figure refines all elements of a design (Refine-All setting), while the bottom one refines a single text box (Refine-Text setting). We see that our approach is able to automatically improve the…
Figure 5
Figure 5. Figure 5: These occlusion-based sensitivity maps show the loca [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Examples where Design-o-meter fails to achieve the optimal refinement. We showcase failure cases of Design-o-meter in [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: We add more qualitative results here. The input design and its corresponding refined output along with the scores are shown. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: We add more qualitative results here. The input design and its corresponding refined output along with the scores are shown. [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: We show the perturbations that we apply to input design, for creating [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 51 canonical work pages

  1. [1]

    Masked siamese networks for label-efficient learning

    Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bo- janowski, Florian Bordes, Pascal Vincent, Armand Joulin, Mike Rabbat, and Nicolas Ballas. Masked siamese networks for label-efficient learning. InEuropean Conference on Com- puter Vision, pages 456–473. Springer, 2022. 2

    work page 2022

  2. [2]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization, 2016. 8

    work page 2016

  3. [3]

    Vi- creg: Variance-invariance-covariance regularization for self- supervised learning

    Adrien Bardes, Jean Ponce, and Yann LeCun. Vi- creg: Variance-invariance-covariance regularization for self- supervised learning. arXiv preprint arXiv:2105.04906, 2021. 2

    arXiv 2021

  4. [4]

    Computational modeling and experimental investigation of effects of compositional ele- ments on interface and design aesthetics

    Michael Bauerly and Yili Liu. Computational modeling and experimental investigation of effects of compositional ele- ments on interface and design aesthetics. International jour- nal of human-computer studies, 64(8):670–682, 2006. 2

    work page 2006

  5. [5]

    Signature verification using a” siamese” time delay neural network

    Jane Bromley, Isabelle Guyon, Yann LeCun, Eduard S¨ackinger, and Roopak Shah. Signature verification using a” siamese” time delay neural network. Advances in neural information processing systems, 6, 1993. 2

    work page 1993

  6. [6]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In Pro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 9650–9660, 2021. 2

    2021

  7. [7]

    Lay- outdm: Transformer-based diffusion model for layout gener- ation, 2023

    Shang Chai, Liansheng Zhuang, and Fengying Yan. Lay- outdm: Transformer-based diffusion model for layout gener- ation, 2023. 1

    work page 2023

  8. [8]

    Towards aligned lay- out generation via diffusion model with aesthetic constraints,

    Jian Chen, Ruiyi Zhang, Yufan Zhou, Rajiv Jain, Zhiqiang Xu, Ryan Rossi, and Changyou Chen. Towards aligned lay- out generation via diffusion model with aesthetic constraints,

    work page

Show all 64 references
  1. [9]

    Graphic design with large multimodal model

    Yutao Cheng, Zhao Zhang, Maoke Yang, Hui Nie, Chunyuan Li, Xinglong Wu, and Jie Shao. Graphic design with large multimodal model. arXiv preprint arXiv:2404.14368, 2024. 3, 7

    2024 arXiv

  2. [10]

    Out of time: auto- mated lip sync in the wild

    Joon Son Chung and Andrew Zisserman. Out of time: auto- mated lip sync in the wild. In Asian conference on computer vision, pages 251–263. Springer, 2016. 2

    2016

  3. [11]

    An evolution- ary many-objective optimization algorithm using reference- point-based nondominated sorting approach, part i: solving problems with box constraints

    Kalyanmoy Deb and Himanshu Jain. An evolution- ary many-objective optimization algorithm using reference- point-based nondominated sorting approach, part i: solving problems with box constraints. IEEE transactions on evolu- tionary computation, 18(4):577–601, 2013. 5

    2013

  4. [12]

    K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan. A fast and elitist multiobjective genetic algorithm: Nsga-ii. IEEE Transactions on Evolutionary Computation , 6(2):182–197,

  5. [13]

    A fast and elitist multiobjective genetic algo- rithm: Nsga-ii

    Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and TAMT Meyarivan. A fast and elitist multiobjective genetic algo- rithm: Nsga-ii. IEEE transactions on evolutionary computa- tion, 6(2):182–197, 2002. 5

    2002

  6. [14]

    Webthetics: quantifying webpage aesthetics with deep learning

    Qi Dou, Xianjun Sam Zheng, Tongfang Sun, and Pheng-Ann Heng. Webthetics: quantifying webpage aesthetics with deep learning. International Journal of Human-Computer Studies, 124:56–66, 2019. 2, 4

    2019

  7. [15]

    Adaptive sub- gradient methods for online learning and stochastic opti- mization

    John Duchi, Elad Hazan, and Yoram Singer. Adaptive sub- gradient methods for online learning and stochastic opti- mization. Journal of machine learning research, 12(7), 2011. 6

    2011

  8. [16]

    Emotionally enhanced talking face generation

    Sahil Goyal, Shagun Uppal, Sarthak Bhagat, Yi Yu, Yifang Yin, and Rajiv Ratn Shah. Emotionally enhanced talking face generation. arXiv preprint arXiv:2303.11548, 2023. 2

    2023 arXiv

  9. [17]

    Can gpts eval- uate graphic design based on design principles?, 2024

    Daichi Haraguchi, Naoto Inoue, Wataru Shimoda, Hayato Mitani, Seiichi Uchida, and Kota Yamaguchi. Can gpts eval- uate graphic design based on design principles?, 2024. 7

    2024

  10. [18]

    Aesthetic measures for automated document layout

    Steven J Harrington, J Fernando Naveda, Rhys Price Jones, Paul Roetling, and Nishant Thakkar. Aesthetic measures for automated document layout. In Proceedings of the 2004 ACM symposium on Document engineering, pages 109–111,

    2004

  11. [19]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. Advances in neural information processing systems , 30, 2017. 1

    2017

  12. [20]

    Retrieval-augmented layout transformer for content-aware layout generation, 2024

    Daichi Horita, Naoto Inoue, Kotaro Kikuchi, Kota Yam- aguchi, and Kiyoharu Aizawa. Retrieval-augmented layout transformer for content-aware layout generation, 2024. 1

    2024

  13. [21]

    Posterlayout: A new benchmark and approach for content-aware visual-textual presentation layout

    Hsiao Yuan Hsu, Xiangteng He, Yuxin Peng, Hao Kong, and Qing Zhang. Posterlayout: A new benchmark and approach for content-aware visual-textual presentation layout. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6018–6026, 2023. 1

    2023

  14. [22]

    Towards flexible multi-modal document models, 2023

    Naoto Inoue, Kotaro Kikuchi, Edgar Simo-Serra, Mayu Otani, and Kota Yamaguchi. Towards flexible multi-modal document models, 2023. 1, 3, 6, 7

    2023

  15. [23]

    Opencole: Towards reproducible automatic graphic design generation, 2024

    Naoto Inoue, Kento Masui, Wataru Shimoda, and Kota Yam- aguchi. Opencole: Towards reproducible automatic graphic design generation, 2024. 1

    2024

  16. [24]

    Batch normalization: Accelerating deep network training by reducing internal co- variate shift, 2015

    Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal co- variate shift, 2015. 7, 8

    2015

  17. [25]

    Performance comparison of nsga-ii and nsga-iii on various many-objective test problems

    Hisao Ishibuchi, Ryo Imada, Yu Setoguchi, and Yusuke Nojima. Performance comparison of nsga-ii and nsga-iii on various many-objective test problems. In 2016 IEEE Congress on Evolutionary Computation (CEC), pages 3045–

    2016

  18. [26]

    Cole: A hierarchical generation frame- work for graphic design

    Peidong Jia, Chenxuan Li, Zeyu Liu, Yichao Shen, Xingru Chen, Yuhui Yuan, Yinglin Zheng, Dong Chen, Ji Li, Xi- aodong Xie, et al. Cole: A hierarchical generation frame- work for graphic design. arXiv preprint arXiv:2311.16974,

    arXiv

  19. [27]

    Cole: A hierarchical generation framework for multi- layered and editable graphic design, 2024

    Peidong Jia, Chenxuan Li, Yuhui Yuan, Zeyu Liu, Yichao Shen, Bohan Chen, Xingru Chen, Yinglin Zheng, Dong Chen, Ji Li, Xiaodong Xie, Shanghang Zhang, and Baining Guo. Cole: A hierarchical generation framework for multi- layered and editable graphic design, 2024. 1, 6, 7

    2024

  20. [28]

    A review on genetic algorithm: past, present, and future

    Sourabh Katoch, Sumit Singh Chauhan, and Vijay Kumar. A review on genetic algorithm: past, present, and future. Mul- timedia tools and applications, 80:8091–8126, 2021. 3

    2021

  21. [29]

    Photo aesthetics ranking network with at- tributes and content adaptation

    Shu Kong, Xiaohui Shen, Zhe Lin, Radomir Mech, and Charless Fowlkes. Photo aesthetics ranking network with at- tributes and content adaptation. In Computer Vision–ECCV 9 2016: 14th European Conference, Amsterdam, The Nether- lands, October 11–14, 2016, Proceedings, Part I 14 ,...

    2016

  22. [30]

    Aesthetics++: Refining graphic designs by exploring design principles and human preference

    Wenyuan Kong, Zhaoyun Jiang, Shizhao Sun, Zhuoning Guo, Weiwei Cui, Ting Liu, Jian-Guang Lou, and Dongmei Zhang. Aesthetics++: Refining graphic designs by exploring design principles and human preference. IEEE Transactions on Visualization and Computer Graphics, 2022. 2, 3

    2022

  23. [31]

    Har- monious textual layout generation over natural images via deep aesthetics learning

    Chenhui Li, Peiying Zhang, and Changbo Wang. Har- monious textual layout generation over natural images via deep aesthetics learning. IEEE Transactions on Multimedia, 24:3416–3428, 2021. 6

    2021

  24. [32]

    Comparison between moea/d and nsga-iii on a set of novel many and multi-objective bench- mark problems with challenging difficulties

    Hui Li, Kalyanmoy Deb, Qingfu Zhang, Ponnuthurai N Sug- anthan, and Lei Chen. Comparison between moea/d and nsga-iii on a set of novel many and multi-objective bench- mark problems with challenging difficulties. Swarm and Evolutionary Computation, 46:104–117, 2019. 5

    2019

  25. [33]

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

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Interna- tional conference on machine learning, pages 12888–12900. PMLR, 2022. 4

    2022

  26. [34]

    Attribute-conditioned lay- out gan for automatic graphic design

    Jianan Li, Jimei Yang, Jianming Zhang, Chang Liu, Christina Wang, and Tingfa Xu. Attribute-conditioned lay- out gan for automatic graphic design. IEEE Transactions on Visualization and Computer Graphics , 27(10):4039–4048,

  27. [35]

    Llava-next: Im- proved reasoning, ocr, and world knowledge, January 2024

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. Llava-next: Im- proved reasoning, ocr, and world knowledge, January 2024. 3, 6

    2024

  28. [36]

    Another look at a model for evaluating interface aesthetics

    David Chek Ling Ngo and John G Byrne. Another look at a model for evaluating interface aesthetics. Interna- tional Journal of Applied Mathematics and Computer Sci- ence, 11(2):515–535, 2001. 2

    2001

  29. [37]

    A mathematical theory of interface aesthetics

    David Chek Ling Ngo, Lian Seng Teo, and John G Byrne. A mathematical theory of interface aesthetics. In Visual math- ematics, volume 2. Mathematical Institute SASA, 2000. 2

    2000

  30. [38]

    Designscape: Design with interactive layout suggestions

    Peter O’Donovan, Aseem Agarwala, and Aaron Hertzmann. Designscape: Design with interactive layout suggestions. In Proceedings of the 33rd annual ACM conference on human factors in computing systems, pages 1221–1224, 2015. 3

    2015

  31. [39]

    Chatgpt can now see, hear, and speak

    OpenAI. Chatgpt can now see, hear, and speak. https: //openai.com/index/chatgpt-can-now-see- hear-and-speak/, 2023. 6

    2023

  32. [40]

    Hello gpt-4o

    OpenAI. Hello gpt-4o. https : / / openai . com / index/hello-gpt-4o/, 2024. 3, 6

    2024

  33. [41]

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anad- kat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Bal- com, Paul Baltescu, Haiming Bao, Mohammad Bavarian, J...

    2024

  34. [42]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 4

    2023 arXiv

  35. [43]

    Learning layouts for single-pagegraphic designs

    Peter O’Donovan, Aseem Agarwala, and Aaron Hertz- mann. Learning layouts for single-pagegraphic designs. IEEE transactions on visualization and computer graphics , 20(8):1200–1213, 2014. 2, 3

    2014

  36. [44]

    Directing user attention via visual flow on web de- signs

    Xufang Pang, Ying Cao, Rynson WH Lau, and Antoni B Chan. Directing user attention via visual flow on web de- signs. ACM Transactions on Graphics (TOG) , 35(6):1–11,

  37. [45]

    A lip sync expert is all you need for speech to lip generation in the wild

    KR Prajwal, Rudrabha Mukhopadhyay, Vinay P Nambood- iri, and CV Jawahar. A lip sync expert is all you need for speech to lip generation in the wild. In Proceedings of the 28th ACM international conference on multimedia , pages 484–492, 2020. 2

    2020

  38. [46]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  39. [47]

    Ruite: Refining ui layout aesthetics using transformer encoder

    Soliha Rahman, Vinoth Pandian Sermuga Pandian, and Matthias Jarke. Ruite: Refining ui layout aesthetics using transformer encoder. In 26th International Conference on Intelligent User Interfaces-Companion, pages 81–83, 2021. 3

    2021

  40. [48]

    Quantifying vi- sual preferences around the world

    Katharina Reinecke and Krzysztof Z Gajos. Quantifying vi- sual preferences around the world. In Proceedings of the SIGCHI conference on human factors in computing systems, pages 11–20, 2014. 2

    2014

  41. [49]

    Visual explanation for advertising creative workflow

    Shoko Sawada, Tomoyuki Suzuki, Kota Yamaguchi, and Masashi Toyoda. Visual explanation for advertising creative workflow. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, pages 1–8, 2024. 2, 4

    2024

  42. [50]

    Aesthetic measures for document layouts: Operationalization and analysis in the context of marketing brochures

    David Sch ¨olgens, Sven M ¨uller, Christine Bauer, Roman Tilly, and Detlef Schoder. Aesthetic measures for document layouts: Operationalization and analysis in the context of marketing brochures. In Proceedings of the 2016 ACM Sym- posium on Document Engineering, pages 21–30, 2016. 2

    2016

  43. [51]

    Vi- sual layout composer: Image-vector dual diffusion model for design layout generation

    Mohammad Amin Shabani, Zhaowen Wang, Difan Liu, Nanxuan Zhao, Jimei Yang, and Yasutaka Furukawa. Vi- sual layout composer: Image-vector dual diffusion model for design layout generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pag...

    2024

  44. [52]

    Muiltiobjective optimization using nondominated sorting in genetic algo- rithms

    Nidamarthi Srinivas and Kalyanmoy Deb. Muiltiobjective optimization using nondominated sorting in genetic algo- rithms. Evolutionary computation, 2(3):221–248, 1994. 3

    1994

  45. [53]

    Automatic layout generation for graphical de- sign magazines

    Sou Tabata, Hiroki Yoshihara, Haruka Maeda, and Kei Yokoyama. Automatic layout generation for graphical de- sign magazines. In ACM SIGGRAPH 2019 Posters , pages 1–2. 2019. 2

    2019

  46. [54]

    In- stance normalization: The missing ingredient for fast styliza- tion, 2017

    Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. In- stance normalization: The missing ingredient for fast styliza- tion, 2017. 8

    2017

  47. [55]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. 3, 4

    2017

  48. [56]

    A novel webpage layout aesthetic evaluation model for quantifying webpage layout design

    Hongyan Wan, Wanting Ji, Guoqing Wu, Xiaoyun Jia, Xue Zhan, Mengting Yuan, and Ruili Wang. A novel webpage layout aesthetic evaluation model for quantifying webpage layout design. Information Sciences, 576:589–608, 2021. 2, 4

    2021

  49. [57]

    Haohan Weng, Danqing Huang, Yu Qiao, Zheng Hu, Chin- Yew Lin, Tong Zhang, and C. L. Philip Chen. Desigen: A pipeline for controllable design template generation, 2024. 1

    2024

  50. [58]

    Group normalization, 2018

    Yuxin Wu and Kaiming He. Group normalization, 2018. 6, 8

    2018

  51. [59]

    Canvasvae: Learning to generate vector graphic documents

    Kota Yamaguchi. Canvasvae: Learning to generate vector graphic documents. ICCV, 2021. 4, 6, 7

    2021

  52. [60]

    Fontdiffuser: One-shot font genera- tion via denoising diffusion with multi-scale content aggre- gation and style contrastive learning, 2023

    Zhenhua Yang, Dezhi Peng, Yuxin Kong, Yuyi Zhang, Cong Yao, and Lianwen Jin. Fontdiffuser: One-shot font genera- tion via denoising diffusion with multi-scale content aggre- gation and style contrastive learning, 2023. 1

    2023

  53. [61]

    Barlow twins: Self-supervised learning via redundancy reduction

    Jure Zbontar, Li Jing, Ishan Misra, Yann LeCun, and St´ephane Deny. Barlow twins: Self-supervised learning via redundancy reduction. In International Conference on Ma- chine Learning, pages 12310–12320. PMLR, 2021. 2

    2021

  54. [62]

    Towards an evalu- ation of graphical user interfaces aesthetics based on met- rics

    Mathieu Zen and Jean Vanderdonckt. Towards an evalu- ation of graphical user interfaces aesthetics based on met- rics. In 2014 IEEE Eighth International Conference on Re- search Challenges in Information Science (RCIS) , pages 1–

    2014

  55. [63]

    What char- acterizes personalities of graphic designs? ACM Transac- tions on Graphics (TOG), 37(4):1–15, 2018

    Nanxuan Zhao, Ying Cao, and Rynson WH Lau. What char- acterizes personalities of graphic designs? ACM Transac- tions on Graphics (TOG), 37(4):1–15, 2018. 2

    2018

  56. [64]

    good” and “bad

    Wanrong Zhu, Jennifer Healey, Ruiyi Zhang, William Yang Wang, and Tong Sun. Automatic layout planning for visually-rich documents with instruction-following models. arXiv preprint arXiv:2404.15271, 2024. 6 11 A. Additional Ablation Experiments A.1. Alternate Margins for Eqn. 3...

    2024 arXiv

Pith tools

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