pith. sign in

arxiv: 2507.02212 · v2 · submitted 2025-07-03 · 💻 cs.CV · cs.CL· cs.LG

SciGA: A Comprehensive Dataset for Designing Graphical Abstracts in Academic Papers

Takuro Kawada , Shunsuke Kitada , Sota Nemoto , Hitoshi Iyatomi This is my paper

Pith reviewed 2026-05-19 05:55 UTC · model grok-4.3

classification 💻 cs.CV cs.CLcs.LG
keywords graphical abstractsdatasetrecommendationscientific visualizationAI for sciencefigure selectionacademic papers
0
0 comments X

The pith

A dataset of 145,000 papers and 1.14 million figures supports AI tasks for selecting and recommending graphical abstracts.

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

The paper introduces SciGA-145k, a large collection of scientific papers and their figures aimed at helping develop AI tools for designing graphical abstracts. It sets up two recommendation tasks: choosing good figures from within the same paper and finding inspiring examples from other papers. A new metric called CAR is proposed to evaluate these recommendations more accurately by recognizing multiple plausible choices. This setup provides a way to use AI to improve how research findings are visually communicated.

Core claim

By releasing the SciGA-145k dataset and defining intra-GA and inter-GA recommendation tasks along with the Confidence Adjusted top-1 ground truth Ratio metric, the work demonstrates the viability of using machine learning for graphical abstract design support in academic papers.

What carries the argument

SciGA-145k dataset of 145k papers and 1.14M figures, which enables training and evaluation of models for intra-paper and inter-paper graphical abstract recommendation.

Load-bearing premise

That the figures chosen as graphical abstracts in the source papers, plus other figures from those papers, serve as reliable examples of good graphical abstracts.

What would settle it

Conducting a controlled experiment where domain experts rate the quality of AI-recommended figures versus non-recommended ones for use as graphical abstracts would test the practical usefulness of the tasks and metric.

Figures

Figures reproduced from arXiv: 2507.02212 by Hitoshi Iyatomi, Shunsuke Kitada, Sota Nemoto, Takuro Kawada.

Figure 1
Figure 1. Figure 1: Example GAs and their annotations in our SciGA-145k. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples of Intra-GA Recommendation results demonstrating the intuition behind CAR@ [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the embeddings, with colors represent [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative examples of Intra-GA Recommendation results obtained by the best-performing baseline [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: User study results for Inter-GA Recommendation. (a) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of research fields of papers included in SciGA-145k. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Statistical overview of SciGA-145k across top-level arXiv categories. (a) Average token length of titles, (b) average token length [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Overview of the contrastive learning framework for method (iv) Abs2Fig w/cap applied to (a) Intra-GA Recommendation and (b) [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of CAR@5 scores across test queries for different values of [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Mean and standard deviation of CAR@5 scores across test queries for different values of [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of CAR@5 scores across individual queries for the best-performing models in each Intra-GA recommendation [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative examples of Intra-GA Recommendation results obtained by the best-performing model (Long-CLIP within method [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Qualitative examples of Intra-GA Recommendation results obtained by the best-performing baseline. [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Examples of Inter-GA recommendation results obtained by different methods. Pink-highlighted research fields or keywords [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Screenshot of the questionnaire used in the user study. (a) The introductory section of the questionnaire, asking participants about [PITH_FULL_IMAGE:figures/full_fig_p021_15.png] view at source ↗
read the original abstract

Graphical Abstracts (GAs) play a crucial role in visually conveying the key findings of scientific papers. Although recent research increasingly incorporates visual materials such as Figure 1 as de facto GAs, their potential to enhance scientific communication remains largely unexplored. Designing effective GAs requires advanced visualization skills, hindering their widespread adoption. To tackle these challenges, we introduce SciGA-145k, a large-scale dataset comprising approximately 145,000 scientific papers and 1.14 million figures, specifically designed to support GA selection and recommendation, and to facilitate research in automated GA generation. As a preliminary step toward GA design support, we define two tasks: 1) Intra-GA Recommendation, identifying figures within a given paper well-suited as GAs, and 2) Inter-GA Recommendation, retrieving GAs from other papers to inspire new GA designs. Furthermore, we propose Confidence Adjusted top-1 ground truth Ratio (CAR), a novel recommendation metric for fine-grained analysis of model behavior. CAR addresses limitations of traditional rank-based metrics by considering that not only an explicitly labeled GA but also other in-paper figures may plausibly serve as GAs. Benchmark results demonstrate the viability of our tasks and the effectiveness of CAR. Collectively, these establish a foundation for advancing scientific communication within AI for Science.

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

1 major / 2 minor

Summary. The paper introduces SciGA-145k, a dataset of approximately 145,000 scientific papers and 1.14 million figures, to support research on graphical abstract (GA) design. It defines two tasks (Intra-GA Recommendation within a paper and Inter-GA Recommendation across papers), proposes the Confidence Adjusted top-1 ground truth Ratio (CAR) metric to account for plausible alternative GAs, and reports benchmark results claiming to demonstrate task viability and CAR effectiveness for advancing AI-assisted scientific communication.

Significance. If the results hold, the work supplies a large-scale public resource that could accelerate development of automated tools for visual scientific summarization. The scale of the dataset and the introduction of CAR—which adjusts standard ranking metrics to credit other in-paper figures as potentially suitable GAs—are concrete strengths that address practical evaluation challenges in recommendation settings.

major comments (1)
  1. [§3] §3 (Dataset Construction): Positive labels for effective GAs are defined solely via figures explicitly presented as GAs in the collected papers together with other in-paper figures. No independent expert validation, ratings on design criteria (clarity, information density, visual hierarchy), or inter-annotator agreement is reported. This assumption is load-bearing for the central claim in §5 that benchmarks demonstrate task viability and CAR effectiveness; without measurable superiority of the labeled GAs, the reported numbers risk reflecting dataset artifacts rather than genuine progress on identifying well-suited GAs.
minor comments (2)
  1. [Abstract] The abstract states benchmark results demonstrate viability but does not include even a single key performance figure (e.g., CAR score or top-1 accuracy), which would help readers quickly gauge the strength of the empirical support.
  2. [§4] Notation for the CAR adjustment formula would be clearer if presented as an explicit equation in the main text rather than referenced only descriptively.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback and for recognizing the potential value of the SciGA-145k dataset and the CAR metric. We address the single major comment below and describe the revisions we intend to incorporate.

read point-by-point responses

  1. Referee: [§3] §3 (Dataset Construction): Positive labels for effective GAs are defined solely via figures explicitly presented as GAs in the collected papers together with other in-paper figures. No independent expert validation, ratings on design criteria (clarity, information density, visual hierarchy), or inter-annotator agreement is reported. This assumption is load-bearing for the central claim in §5 that benchmarks demonstrate task viability and CAR effectiveness; without measurable superiority of the labeled GAs, the reported numbers risk reflecting dataset artifacts rather than genuine progress on identifying well-suited GAs.

    Authors: We agree that the labeling relies on figures already designated as GAs by the original paper authors, together with the remaining in-paper figures. This choice was deliberate: it leverages existing publishing conventions at scale (145k papers) rather than introducing new subjective annotations that would be costly and difficult to scale. The Intra-GA task is explicitly formulated to treat multiple in-paper figures as potentially suitable, which is why CAR was developed to credit plausible alternatives instead of penalizing them. The reported benchmarks therefore demonstrate that models can be trained to recover the author-chosen GA and to surface other viable candidates, providing an initial proof of concept for the tasks. Nevertheless, we recognize that the absence of independent expert ratings on criteria such as clarity or visual hierarchy constitutes a genuine limitation for stronger claims about label quality. In the revised manuscript we will (i) add an explicit limitations paragraph in §3 detailing this assumption and its implications, (ii) include a forward-looking statement on planned expert validation studies, and (iii) temper the language in §5 to emphasize that the results establish task viability under the current labeling rather than absolute superiority of the labeled GAs. These changes will make the evidential basis clearer without altering the dataset or core experiments. revision: partial

Circularity Check

0 steps flagged

No significant circularity; dataset, tasks, and CAR metric are explicitly defined without reduction to inputs by construction.

full rationale

The paper centers on releasing SciGA-145k and defining Intra-GA and Inter-GA recommendation tasks plus the CAR metric. No equations, derivations, or predictions are presented that reduce by construction to fitted parameters, self-referential quantities, or self-citation chains. Ground truth is explicitly constructed from paper-provided GAs and in-paper figures, and benchmarks are empirical evaluations on this dataset; the contribution remains self-contained with independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The work rests on the domain assumption that existing paper figures can serve as proxies for graphical abstract quality and that recommendation tasks meaningfully advance GA design support.

axioms (1)
  • domain assumption Some figures within scientific papers are suitable to function as graphical abstracts
    Invoked when defining intra-GA recommendation and when treating non-labeled figures as plausible ground truth for CAR.
invented entities (1)
  • CAR metric no independent evidence
    purpose: To provide fine-grained evaluation that credits models for selecting any plausible GA rather than only the single labeled one
    Newly defined metric introduced to address limitations of traditional rank-based metrics.

pith-pipeline@v0.9.0 · 5774 in / 1264 out tokens · 34019 ms · 2026-05-19T05:55:53.807146+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages · 2 internal anchors

  1. [1]

    ResearchAgent: Iterative Research Idea Generation over Scientific Literature with Large Language Models, 2024

    Jinheon Baek, Sujay Kumar Jauhar, Silviu Cucerzan, and Sung Ju Hwang. ResearchAgent: Iterative Research Idea Generation over Scientific Literature with Large Language Models, 2024. https://doi.org/10.48550/arXiv.2404.07738. 1

    work page doi:10.48550/arxiv.2404.07738 2024

  2. [2]

    METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments

    Satanjeev Banerjee and Alon Lavie. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments. In ACL, 2005. 5, 7, 8

    work page 2005

  3. [3]

    Graphical abstracts are associated with greater Altmetric attention scores, but not citations, in sport science

    Hunter Bennett and Flynn Slattery. Graphical abstracts are associated with greater Altmetric attention scores, but not citations, in sport science. Scientometrics, 128:3793–3804,

    work page

  4. [4]

    Buchanan and Edward A

    Bruce G. Buchanan and Edward A. Feigenbaum. Dendral and meta-dendral: Their applications dimension. Artificial Intelligence, 11(1):5–12, 1978. 1

    work page 1978

  5. [5]

    Burges, Tal Shaked, Erin Renshaw, Ari Lazier, Matt Deeds, Nicole Hamilton, and Greg Hullender

    Christopher J.C. Burges, Tal Shaked, Erin Renshaw, Ari Lazier, Matt Deeds, Nicole Hamilton, and Greg Hullender. Learning to rank using gradient descent. In ICML, 2005. 4

    work page 2005

  6. [6]

    S. J. Chapman, R. C. Grossman, M. E. B. FitzPatrick, and R. R. W. Brady. Randomized controlled trial of plain English and visual abstracts for disseminating surgical research via social media. British Journal of Surgery, 106(12):1611–1616,

    work page

  7. [7]

    Reproducible scal- ing laws for contrastive language-image learning

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuh- mann, Ludwig Schmidt, and Jenia Jitsev. Reproducible scal- ing laws for contrastive language-image learning. In CVPR,

    work page

  8. [8]

    Automatic Evaluation of Machine Translation Quality Using Longest Common Sub- sequence and Skip-Bigram Statistics

    Franz Josef Och Chin-Yew Lin. Automatic Evaluation of Machine Translation Quality Using Longest Common Sub- sequence and Skip-Bigram Statistics. In ACL, 2004. 5, 7, 8

    work page 2004

  9. [9]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. In ICLR, 2021. 2, 5, 7, 4

    work page 2021

  10. [10]

    DOC2PPT: Automatic Presentation Slides Generation from Scientific Documents

    Tsu-Jui Fu, William Yang Wang, Daniel McDuff, and Yale Song. DOC2PPT: Automatic Presentation Slides Generation from Scientific Documents. In AAAI, 2022. 1, 3

    work page 2022

  11. [11]

    ar5iv:04.2024 dataset, an HTML5 conversion of arXiv.org, 2024

    Deyan Ginev. ar5iv:04.2024 dataset, an HTML5 conversion of arXiv.org, 2024. SIGMathLing – Special Interest Group on Math Linguistics. 4

    work page 2024

  12. [12]

    CLIPScore: A Reference-free Evaluation Metric for Image Captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. CLIPScore: A Reference-free Evaluation Metric for Image Captioning. In EMNLP, 2021. 5

    work page 2021

  13. [13]

    Hoffberg, Joe Huggins, Audrey Cobb, Jeri E

    Adam S. Hoffberg, Joe Huggins, Audrey Cobb, Jeri E. Forster, and Nazanin H. Bahraini. Beyond Journals—Visual Abstracts Promote Wider Suicide Prevention Research Dissemination and Engagement: A Randomized Crossover Trial. Frontiers in Research Metrics and Analytics, 5, 2020. 2, 3

    work page 2020

  14. [14]

    Martin, Calvin H

    Simon Huang, Lynsey J. Martin, Calvin H. Yeh, Alvin Chin, Heather Murray, William B. Sanderson, Rohit Mohindra, Teresa M. Chan, and Brent Thoma. The effect of an info- graphic promotion on research dissemination and readership: A randomized controlled trial. Canadian Journal of Emer- gency Medicine, 20(6):826—-833, 2018. 1

    work page 2018

  15. [15]

    Ibrahim, Keith D

    Andrew M. Ibrahim, Keith D. Lillemoe, Mary E. Klingen- smith, and Justin B. Dimick. Visual Abstracts to Dissemi- nate Research on Social Media A Prospective, Case-control Crossover Study. Annals of Surgery, 266(6):46–48, 2017. 2, 3

    work page 2017

  16. [16]

    Attract readers with a graphical abstract – The latest clickbait

    Madhan Jeyaraman and Raju Vaishya. Attract readers with a graphical abstract – The latest clickbait. Journal of Or- thopaedics. Journal of Orthopaedics , 38(1):30–31, 2023. 2, 3

    work page 2023

  17. [17]

    Madhan Jeyaraman, Harish V . K. Ratna, Naveen Jeyara- man, Nicola Maffulli, Filippo Migliorini, Arulkumar Nal- lakumarasamy, and Sankalp Yadav. Graphical Abstract in Scientific Research. Cureus, 15(9), 2023. 2, 3, 4

    work page 2023

  18. [18]

    Seeing Is Believing: The Effect of Graphical Abstracts on Citations and Social Media Exposure in Gastroenterology & Hepatology Journals

    Yohan Kim, Ji-Eun Lee, Jeong-Ju Yoo, Eun-Ae Jung, Sang Gyune Kim, and Young Seok Kim. Seeing Is Believing: The Effect of Graphical Abstracts on Citations and Social Media Exposure in Gastroenterology & Hepatology Journals. Journal of Korean Medical Science, 37, 2022. 3

    work page 2022

  19. [19]

    Krukowski and Carly M

    Rebecca A. Krukowski and Carly M. Goldstein. The potential for graphical abstracts to enhance science communication. Transl Behav Med, 13(12):891–895, 2023. 3

    work page 2023

  20. [20]

    Kunze, Amar Vadhera, Ritika Purbeyc, Harsh Singh, Gregory S

    Kyle N. Kunze, Amar Vadhera, Ritika Purbeyc, Harsh Singh, Gregory S. Kazarian, and Jorge Chahla. Infographics are more effective at increasing social media attention in comparison with original research articles: An altmetrics-based analysis. Canadian Journal of Emergency Medicine, 37(8):2591–2597,

    work page

  21. [21]

    The current state of graphical abstracts and how to create good graphical abstracts

    Jieun Lee and Jeong-Ju Yoo. The current state of graphical abstracts and how to create good graphical abstracts. Science Editing, 10(1):19–26, 2023. 2, 3

    work page 2023

  22. [22]

    Douglas B. Lenat. Automated Theory Formation in Mathe- matics. In IJCAI, 1977. 1

    work page 1977

  23. [23]

    Lenat and John Seely Brown

    Douglas B. Lenat and John Seely Brown. Why am and eurisko appear to work. In AAAI, 1983. 1

    work page 1983

  24. [24]

    BLIP- 2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. BLIP- 2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models. In ICML, 2023. 6, 7, 8, 4

    work page 2023

  25. [25]

    Multimodal ArXiv: A Dataset for Improving Scientific Comprehension of Large Vision- Language Models

    Lei Li, Yuqi Wang, Runxin Xu, Peiyi Wang, Xiachong Feng, Lingpeng Kong, and Qi Liu. Multimodal ArXiv: A Dataset for Improving Scientific Comprehension of Large Vision- Language Models. In ACL, 2024. 3

    work page 2024

  26. [26]

    Wilson, Woosang Lim, and William Yang Wang

    Zekun Li, Xianjun Yang, Kyuri Choi, Wanrong Zhu, Ryan Hsieh, HyeonJung Kim, Jin Hyuk Lim, Sungy- oung Ji, Byungju Lee, Xifeng Yan, Linda Ruth Petzold, Stephen D. Wilson, Woosang Lim, and William Yang Wang. MMSci: A Dataset for Graduate-Level Multi- Discipline Multimodal Scientific Understanding, 2024. https://doi.org/10.48550/arXiv.2407.04903. 3

    work page doi:10.48550/arxiv.2407.04903 2024

  27. [27]

    Swin Transformer V2: Scaling Up Capacity and Resolution

    Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, and Baining Guo. Swin Transformer V2: Scaling Up Capacity and Resolution. In CVPR, 2022. 2, 5, 7, 4

    work page 2022

  28. [28]

    S2ORC: The Semantic Scholar Open Re- search Corpus

    Kyle Lo, Lucy Lu Wang, Mark Neumann, Rodney Kinney, and Daniel Weld. S2ORC: The Semantic Scholar Open Re- search Corpus. In ACL, 2020. 3

    work page 2020

  29. [29]

    The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery

    Chris Lu, Cong Lu, Robert Lange, Jakob Foerste, Jeff Clune, and David Ha. The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery, 2024. https://doi.org/10.48550/arXiv.2408.06292. 1

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2408.06292 2024

  30. [30]

    UMAP: Uniform Manifold Approximation and Projection

    Leland McInnes, John Healy, Nathaniel Saul, and Lukas Grossberger. UMAP: Uniform Manifold Approximation and Projection. The Journal of Open Source Software, 3(29):861,

    work page

  31. [31]

    Lennart Meincke, Karan Girotra, Gideon Nave, Christian Terwiesch, and Karl T. Ulrich. Using Large Language Models for Idea Generation in Innovation. SSRN Electronic Journal,

    work page

  32. [32]

    Schoenholz, Mu- ratahan Aykol, Gowoon Cheon, and Ekin Dogus Cubuk

    Amil Merchant, Simon Batzner, Samuel S. Schoenholz, Mu- ratahan Aykol, Gowoon Cheon, and Ekin Dogus Cubuk. Scal- ing deep learning for materials discovery. Nature, 624:80–85,

    work page

  33. [33]

    The Role of Visual Abstracts in the Dissemination of Medical Research

    Beverley C Millar and Michelle Lim. The Role of Visual Abstracts in the Dissemination of Medical Research. Ulster Medical Journal, 91(2):67–78, 2022. 3, 4

    work page 2022

  34. [34]

    Pyzer-Knapp, Jed W

    Edward O. Pyzer-Knapp, Jed W. Pitera, Peter W. J. Staar, Seiji Takeda, Teodoro Laino, Daniel P. Sanders, James Sex- ton, John R. Smith, and Alessandro Curioni. Accelerating materials discovery using artificial intelligence, high perfor- mance computing and robotics. npj Computational Materials, 8(84), 2022. 1

    work page 2022

  35. [35]

    Learning Transferable Visual Models From Natural Language Supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning Transferable Visual Models From Natural Language Supervision. In ICML, 2022. 2, 4, 7, 8

    work page 2022

  36. [36]

    Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

    Nils Reimers and Iryna Gurevych. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In EMNLP,

    work page

  37. [37]

    Robertson, Steve Walker, Susan Jones, Micheline Hancock-Beaulieu, and Mike Gatford

    Stephen E. Robertson, Steve Walker, Susan Jones, Micheline Hancock-Beaulieu, and Mike Gatford. Okapi at trec-3. In TREC-3, 1994. 5, 7, 8

    work page 1994

  38. [38]

    Rodriguez, David Vazquez, Issam Laradji, Marco Pedersoli, and Pau Rodriguez

    Juan A. Rodriguez, David Vazquez, Issam Laradji, Marco Pedersoli, and Pau Rodriguez. FigGen: Text to Scientific Figure Generation. In ICLR, 2023. 1, 3

    work page 2023

  39. [39]

    Rodriguez, David Vazquez, Issam Laradji, Marco Pedersoli, and Pau Rodriguez

    Juan A. Rodriguez, David Vazquez, Issam Laradji, Marco Pedersoli, and Pau Rodriguez. OCR-VQGAN: Taming Text- within-Image Generation. In WACV, 2023. 3

    work page 2023

  40. [40]

    unarXive 2022: All arXiv Publications Pre-Processed for NLP, Includ- ing Structured Full-Text and Citation Network

    Tarek Saier, Johan Krause, and Michael F ¨arber. unarXive 2022: All arXiv Publications Pre-Processed for NLP, Includ- ing Structured Full-Text and Citation Network. In JCDL,

    work page 2022

  41. [41]

    Szymanski, Bernardus Rendy, Yuxing Fei, Rishi E

    Nathan J. Szymanski, Bernardus Rendy, Yuxing Fei, Rishi E. Kumar, Tanjin He, David Milsted, Matthew J. McDermott, Max Gallant, Ekin Dogus Cubuk, Amil Merchant, Haegyeom Kim, Anubhav Jain, Christopher J. Bartel, Kristin Persson, Yan Zeng, and Gerbrand Ceder. An autonomous laboratory for the accelerated synthesis of novel materials. Nature, 624: 86–91, 2023. 1

    work page 2023

  42. [42]

    Mingxing Tan and Quoc V . Le. Efficientnetv2: Smaller Mod- els and Faster Training. In ICML, 2021. 5, 7, 4

    work page 2021

  43. [43]

    SciPost- Layout: A Dataset for Layout Analysis and Layout Genera- tion of Scientific Posters

    Shohei Tanaka, Hao Wang, and Yoshitaka Ushiku. SciPost- Layout: A Dataset for Layout Analysis and Layout Genera- tion of Scientific Posters. In BMVC, 2024. 1, 3

    work page 2024

  44. [44]

    Learning to Generate Posters of Scientific Papers by Probabilistic Graphical Models

    Yu ting Qiang, Yanwei Fu, Xiao Yu, Yanwen Guo, Zhi-Hua Zhou, and Leonid Sigal. Learning to Generate Posters of Scientific Papers by Probabilistic Graphical Models. Journal of Computer Science and Technology, 34:155–169, 2019. 3

    work page 2019

  45. [45]

    Representation Learning with Contrastive Predictive Coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Repre- sentation learning with contrastive predictive coding, 2018. https://doi.org/10.48550/arXiv.1807.03748. 6

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1807.03748 2018

  46. [46]

    Lawrence Zitnick, and Devi Parikh

    Ramakrishna Vedantam, C. Lawrence Zitnick, and Devi Parikh. Cider: Consensus-based image description evalu- ation. In CVPR, 2015. 5, 7, 8

    work page 2015

  47. [47]

    OverleafCopilot: Empowering Aca- demic Writing in Overleaf with Large Language Models,

    Haomin Wen, Zhenjie Wei, Yan Lin, Jiyuan Wang, Yuxuan Liang, and Huaiyu Wan. OverleafCopilot: Empowering Aca- demic Writing in Overleaf with Large Language Models,

    work page

  48. [48]

    https://doi.org/10.48550/arXiv.2403.09733. 1

    work page doi:10.48550/arxiv.2403.09733

  49. [49]

    Con- vNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders

    Sanghyun Woo, Shoubhik Debnath, Ronghang Hu, Xinlei Chen, Zhuang Liu, In So Kweon, and Saining Xie. Con- vNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders. In CVPR, 2023. 2, 5, 7, 4

    work page 2023

  50. [50]

    Automatic Paper Summary Generation from Visual and Textual Information

    Shintaro Yamamoto, Yoshihiro Fukuhara, Ryota Suzuki, Shi- geo Morishima, and Hirokatsu Kataoka. Automatic Paper Summary Generation from Visual and Textual Information. In ICMV, 2018. 3, 5

    work page 2018

  51. [51]

    Verbal and visual resources in graphical abstracts: Analyzing patterns of knowledge pre- sentation in digital genres

    Ma Yuanyuan and Jiang Kevin. Verbal and visual resources in graphical abstracts: Analyzing patterns of knowledge pre- sentation in digital genres. Ib´erica, 46:129–154, 2023. 4

    work page 2023

  52. [52]

    X2-VLM: All-in-One Pre- Trained Model for Vision-Language Tasks.IEEE transactions on pattern analysis and machine intelligence , 46(5):3156– 3168, 2023

    Yan Zeng, Xinsong Zhang, Hang Li, Jiawei Wang, Jipeng Zhang, and Wangchunshu Zhou. X2-VLM: All-in-One Pre- Trained Model for Vision-Language Tasks.IEEE transactions on pattern analysis and machine intelligence , 46(5):3156– 3168, 2023. 6, 7, 8, 4

    work page 2023

  53. [53]

    Long-CLIP: Unlocking the Long-Text Capability of CLIP

    Beichen Zhang, Pan Zhang, Xiaoyi Dong, Yuhang Zang, and Jiaqi Wang. Long-CLIP: Unlocking the Long-Text Capability of CLIP. In ECCV, 2024. 2, 6, 7, 8, 4

    work page 2024

  54. [54]

    Wein- berger, and Yoav Artzi

    Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Wein- berger, and Yoav Artzi. BERTScore: Evaluating Text Genera- tion with BERT. In ICLR, 2020. 5, 7, 8, 4

    work page 2020

  55. [55]

    ID": "2401.13641

    Hao Zheng, Xinyan Guan, Hao Kong, Jia Zheng, Weix- iang Zhou, Hongyu Lin, Yaojie Lu, Ben He, Xian- pei Han, and Le Sun. PPTAgent: Generating and Evaluating Presentations Beyond Text-to-Slides, 2025. https://doi.org/10.48550/arXiv.2501.03936. 3 A. Dataset Structure The textual data and associated metadata of SciGA-145k are provided in JSON format, as illus...

    work page doi:10.48550/arxiv.2501.03936 2025