Pith. sign in

REVIEW 4 major objections 5 minor 47 references

GVR-Coder turns long professional texts into coherent, editable SVG diagrams by combining layout-aware curriculum training, dual rendering rewards, and a generate-verify-repair loop.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 18:40 UTC pith:L6GUW3CQ

load-bearing objection Solid office text-to-SVG systems paper with a real domain dataset; headline wins over Gemini are partly Gemini-shaped, but the stack and ablations still hold up. the 4 major comments →

arxiv 2607.28073 v1 pith:L6GUW3CQ submitted 2026-07-30 cs.LG cs.CV

GVR-Coder: A Visual-Feedback Framework for Structured SVG Generation in Complex Document and Meeting Scenarios

Yiming Xu , Jihua Kang , Chunsai Du , Qifan Zhang , Wangqiu Zhou , Yiting Wu , Tianqi Li , Qi Song This is my paper
classification cs.LG cs.CV
keywords Large Language ModelsScalable Vector GraphicsText-to-SVGReinforcement LearningVisual FeedbackDocument DiagramsCurriculum LearningMulti-Agent Repair
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Dense meeting notes and office documents hide their logic under long prose, raising cognitive load for anyone who must extract structure. This paper claims Scalable Vector Graphics are the right automatic target because they stay resolution-independent and editable, yet prior text-to-SVG work fails on real office diagrams: suitable data are scarce, models lack layout priors, and nothing ever looks at the rendered picture to fix defects. The authors release DocMeetSVG-100K, a 100,000-pair dataset of document and meeting diagrams graded by difficulty, and introduce GVR-Coder. The system first teaches simple-to-hard structure with explicit layout constraints, then uses reinforcement learning on dual rewards that jointly push visual cleanliness and structural richness, and finally runs a generate-verify-repair agent loop that diagnoses concrete defects and rewrites the code. On held-out meeting, dialogue, and rule-constrained tests, the resulting 14B–32B models reach higher rates of fully usable diagrams than larger closed-source systems while staying factually faithful. A reader who works with long documents cares because the output is something a colleague can skim, edit, and drop into a slide without redrawing.

Core claim

GVR-Coder—curriculum-driven rejection-sampling fine-tuning with injected layout knowledge, reinforcement learning from dual rendering feedback that balances aesthetics against structural complexity, and a generate-verify-repair agent loop—converts lengthy professional text into logically coherent and visually clean SVG diagrams and outperforms competitive baselines, including larger closed-source models, on the authors’ DocMeetSVG benchmarks.

What carries the argument

The GVR-Coder three-stage stack: CRSFT (curriculum rejection-sampling fine-tuning plus explicit layout constraints), RLDRF (GRPO reinforcement learning with a hybrid visual-plus-complexity reward computed on rendered images), and VRL (an inference-time generate-verify-repair multi-agent loop that supplies fine-grained defect attribution and targeted code repair).

Load-bearing premise

The whole pipeline treats one high-tier vision-language model as a reliable oracle for building the training set, labeling aesthetic rewards, scoring final quality, and driving repairs, so that large gains against that same model are taken as genuine diagram quality rather than style matching.

What would settle it

Have independent human designers who never see model names perform a fully blind usability-and-fidelity scoring of the 320-test-set diagrams for GVR-Coder versus the strongest closed-source baseline; if GVR’s acceptance-rate advantage disappears or reverses under pure human judgment, the central claim does not hold.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Meeting minutes and long documents can be turned automatically into editable flowcharts, covers, and topic maps without manual layout work.
  • Open models in the 14B–32B range trained this way can match or exceed much larger closed models on structured office-diagram tasks.
  • Dual rewards that penalize both visual defects and oversimplification keep diagrams both clean and information-dense.
  • Injecting layout constraints plus iterative visual repair becomes a reusable pattern for other code-that-must-look-right generation problems.
  • DocMeetSVG-100K and the aesthetic-plus-factuality protocol become a shared benchmark for logic-rich SVG generation.

Where Pith is reading between the lines

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

  • The same generate-verify-repair loop could transfer to other vector or declarative diagram formats once a stable renderer and a shared defect taxonomy exist.
  • Heavy reliance on a single proprietary verifier at train and test time points to a natural next step: distilling a lightweight open verifier so the loop no longer needs an external oracle at inference.
  • If the reported cognitive-load reduction replicates in larger workplace field studies, the pipeline could become standard post-processing for meeting-transcription and document-authoring tools.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces DocMeetSVG-100K, a 100K Text-to-SVG dataset for document/meeting diagrams, and GVR-Coder, a three-stage framework (curriculum rejection-sampling SFT with layout priors; GRPO with dual visual/complexity rendering rewards; and a generate–verify–repair agent loop) that turns long professional text into structured SVG. On a 320-sample held-out suite spanning Arena dialogues, meeting covers/topics, and rule-constrained docs, staged 14B/32B variants report large gains in average aesthetic score and perfect acceptance rate over traditional SVG models and strong open/closed LLMs, with ablations, cross-VLM re-scoring, designer ratings, a small user ranking study, and latency–quality analysis offered as support.

Significance. If the quality claims hold under more independent evaluation, the work is a solid systems contribution to multimodal code generation for a practically important office setting that existing icon-centric SVG datasets and methods underserve. Concrete strengths include a large, difficulty-stratified domain dataset with type-specific pipelines and partial human audit; an explicit anti-degeneration complexity reward alongside aesthetics; staged ablations (Table 3) and repair-transition analysis (Figure 8); cross-evaluator checks (Table 2); human designer scores (Table 4); and a promised code/data release. The combination of layout-constraint injection, dual rendering RL, and fine-grained repair is a coherent response to the three stated bottlenecks (data, priors, visual feedback).

major comments (4)
  1. [§3, §4.2 Eq. (1), §4.3, §5.1–5.2, Tables 1–2] Load-bearing independence of the Gemini-3-Pro oracle is not adequately established for the headline claim that GVR-Coder “outperforms competitive baselines” (Abstract; §5.2; Table 1). Gemini is used to generate and repair DocMeetSVG-100K candidates (§3), to supply R_visual via six defect types in RLDRF (§4.2, Eq. 1), as the Verify Agent in VRL (§4.3, Eq. 4), and as the primary VLM-as-Judge for As/Ar (§5.1). Table 2 and Fig. 4/human checks reduce pure self-grading but still largely re-score outputs already optimized under Gemini feedback; they do not show that training targets and rewards were built without that preference model. Please either (i) retrain or re-label a substantial reward/data slice with a held-out judge family and re-report Table 1, or (ii) clearly scope the claim as “improves under Gemini-aligned visual criteria” and lead with human and non-Gemini metrics.
  2. [§4.3, §5.2 Table 1, Fig. 5] The strongest Table 1 numbers are for *-VRL variants (e.g., 32B-VRL Ar 100% on Meeting_cover_100, 88.3% on Svg_with_rules_120) compared to single-pass Gemini-3-pro and other LLMs. VRL’s Verify Agent is Gemini-3-Pro and may run up to three repair rounds (§4.3), so this is a multi-agent system with Gemini in the inference loop versus single-pass baselines—not a matched generator comparison. RSFT/RLDRF rows are the fair single-model evidence; VRL should be reported as a system result, with ablations that freeze or replace the verifier (open VLM / rule-only / human feedback) and a same-budget multi-sample or self-refine baseline for Gemini/GPT. Without that, “outperforms Gemini” overstates model quality relative to inference scaffolding.
  3. [§4.2 Eqs. (2)–(3); Appendix Fig. 14–15] R_complex (§4.2, Eq. 2) defines complexity as a count of structural tags (path, circle, line, …) matched to ≥0.8 of a reference SVG. This is a reasonable anti-simplification regularizer (and Fig. 14–15 help), but it is not shown to track logical/topological richness required by the paper’s own motivation (hierarchies, flow connectivity, factual structure). Tag inflation can raise R_complex without improving diagram logic. Please justify the 0.8 threshold, report sensitivity, and add at least one structure-aware check (e.g., node/edge counts by diagram type, connectivity validity, or correlation with factuality) so the dual-reward claim is not carried only by aesthetic VLM scores plus raw tag density.
  4. [§5.1–5.2, Fig. 4, Fig. 7, Table 4] Reliability claims rest on a 320-example test set and modest human studies (200 for judge agreement in Fig. 4; 10 designers in Table 4; 20 users on 50 texts). Given six primary categories and four named benchmarks, please report per-benchmark confidence intervals or bootstrap uncertainty for Ar/As, break down failures by defect type and diagram category, and clarify how much of the test distribution overlaps Gemini’s stylistic priors. The factuality protocol is binary (0/1 on hallucination/omission/logic) and is secondary in the narrative to aesthetic Ar; for “logically coherent” office diagrams, factuality should be tabulated with the same prominence as As/Ar for all main systems, not mainly via the radar summary (Fig. 7).
minor comments (5)
  1. [References; Table 1] Several bibliography entries and in-text model names appear contemporaneous or inconsistently versioned (e.g., GPT-5.x, Gemini-3, GLM-5, Kimi-k2.5). Ensure public, citable snapshots and dates so results are reproducible after model updates.
  2. [§5.5 Fig. 9; Appendix H] Figure 9 and appendix qualitative figures note that high-resolution legible versions are deferred to the supplement; for the main camera-ready, at least one full-resolution case per benchmark should be readable in-page (text in nodes is part of the claim).
  3. [§4.1–4.3; Table 3] Curriculum mixture ratios (100% simple; 85/15; 85/15), rejection width of two candidates, VRL cap of three iterations, and unweighted R_total = R_visual + R_complex are free parameters; a short sensitivity subsection or appendix table would strengthen §5.4.
  4. [Table 1] SVG Validity is listed as N/A for VRL rows in Table 1; briefly explain (post-repair validity assumed?) to avoid implying the metric was dropped when convenient.
  5. [§1 contributions; Table 1; Fig. 3] Minor prose/typos: “protoc al” → protocol (contributions); “A verage” spacing in Table 1 caption; “Staged Curriculum” figure text is dense—consider enlarging reward definitions in Fig. 3.

Circularity Check

2 steps flagged

No definitional tautology in the method math; partial closed-loop dependence where Gemini builds data, labels RL rewards, verifies VRL repairs, and is the primary As/Ar judge—so VRL gains vs Gemini are only partly independent.

specific steps
  1. other [§3 DocMeetSVG-100K; §4.2 Eq.1; §4.3 Verify Agent; §5.1 Automated Evaluation]
    "For each caption, Gemini-3-Pro generates two independent SVG candidates, which are evaluated by a Judge Model... Imperfect samples are further refined by a specialized Repair Model... Valid images are evaluated by a high-tier VLM (Gemini-3-pro) across six dimensions... R_visual = max(0,6−N_errors)/6 ... Leveraging the multimodal capabilities of Gemini-3-Pro, this agent analyzes the original caption, the SVG code P_t, and the rendered image I_t... we adopt Gemini-3-Pro as VLM-as-Judge... Score=max(0,6−N)"

    Training targets, RL aesthetic labels, inference repair stopping rule, and primary accept-ratio metric share one preference model and the same six-defect taxonomy. VRL iterates until that verifier passes; the headline judge then scores the same defect list. This is not X defined as Y algebraically, but reported Ar gains of VRL over single-pass Gemini are partly aligned-by-construction to Gemini’s pass criteria rather than to a fully independent quality oracle. Cross-VLM/human checks mitigate but score outputs already optimized under that loop.

  2. fitted input called prediction [§4.2 Complexity Reward Eq.2; RLDRF on hard samples]
    "R_complex = 1.0 if n_gen ≥ 0.8·n_ref; n_gen/(0.8·n_ref) otherwise. ... Through GRPO optimization on 8,000 hard samples, the model is capable of generating diagrams that meet aesthetic expectations while maintaining structural richness."

    Minor only: complexity is matched to reference tag counts from the Gemini-constructed/repaired ground truth, then models are praised for structural richness comparable to those references. That anchors density to the dataset’s own SVG statistics rather than an external complexity standard. It does not force aesthetic As/Ar by itself (R_visual is separate), so it is a weak fitted-target coupling, not the main claim collapse.

full rationale

This is an empirical systems paper, not a first-principles derivation. CRSFT, the dual-reward definitions (Eqs. 1–3), and the agent loop are operational procedures, not claims that reduce to fitted constants renamed as predictions. There is no self-citation uniqueness theorem, no ansatz smuggled from the authors’ prior math, and no renaming of a known closed-form result. Human audits (30% stratified; 200-sample judge agreement; 10 designers; 20-user ranking), cross-VLM re-scoring (Table 2), and complexity reward grounded in reference tag counts supply external content. The residual issue is methodological closed-loop preference: Gemini-3-Pro is used to generate/repair DocMeetSVG-100K, to supply R_visual in RLDRF, as the VRL Verify Agent, and as primary VLM-as-Judge for headline Ar/As—including comparisons that claim to beat Gemini. That does not make Eq. 1 equal the claim by algebra, but it does make large VRL accept-ratio jumps on the Gemini judge partly expected once repairs run until the same model family says pass. Proportionate score is therefore moderate (not 0, not 6+): central engineering contributions remain independently specified; independence of the strongest “outperforms Gemini” metric is only partially established.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 3 invented entities

The central performance claim rests on synthetic supervision and automatic judges more than on closed-form theory. Load-bearing choices include Gemini as data/reward/judge oracle, hand-set curriculum ratios and complexity threshold, equal-sum hybrid reward, and the assumption that listed layout constraints plus ≤3 repair rounds define ‘office-standard’ diagrams. No new physical entities; free parameters are engineering knobs that shape both training and reported accept rates.

free parameters (5)
  • Curriculum stage mixtures (100% simple; 85/15 medium+simple; 85/15 hard+medium) = 100% / 85%-15% / 85%-15%
    Hand-chosen data schedules in CRSFT that control what the SFT model sees; not derived from a principle.
  • Complexity match threshold 0.8 in R_complex = 0.8
    Eq. 2 awards full complexity credit when n_gen ≥ 0.8 n_ref; the 0.8 cutoff is a design choice that directly shapes RL behavior.
  • R_total = R_visual + R_complex (unweighted sum) = weight 1:1
    Equal weighting of aesthetics vs complexity is chosen, not cross-validated in the main text; it defines the RL objective.
  • Rejection sampling width = 2 candidates; VRL max iterations = 3 = 2 candidates; 3 repair rounds
    Difficulty labels and inference compute/quality trade-off are set by these integers; paper cites plateau beyond 3 repairs.
  • Deductive aesthetic scale max(0, 6 - N_errors)/6 = 6 defect bins, unit deduction
    Six defect categories and unit penalties define both RL rewards and automatic As/Ar metrics.
axioms (4)
  • domain assumption A strong proprietary VLM (Gemini-3-Pro) is an adequate proxy for human judgments of SVG aesthetics, topology, and factual consistency for dataset filtering, RL rewards, and primary benchmarking.
    Invoked throughout §3–§5 and appendix prompts; partially checked via ~90% human agreement on 200 samples and cross-VLM tables, but still foundational.
  • domain assumption Explicit natural-language layout constraints in prompts (collision avoidance, anchor alignment, anti-hallucination) supply usable spatial priors to text-only code models.
    §4.1 knowledge augmentation; ablation credits a modest Ar lift to this injection.
  • ad hoc to paper Tag-count complexity relative to a reference SVG is a valid anti-degeneration regularizer for diagram quality.
    Eq. 2 operationalizes ‘structural richness’; appendix argues it blocks visual-reward oversimplification.
  • standard math Standard supervised fine-tuning, rejection sampling, curriculum learning, and GRPO-style policy optimization are valid learning mechanisms for SVG code.
    Background ML toolkit assumed without new theory proofs.
invented entities (3)
  • DocMeetSVG-100K no independent evidence
    purpose: Provide large-scale text–SVG supervision for logic-rich office/meeting diagrams missing from icon-centric corpora.
    New dataset asset; quality is largely defined by the Gemini generate–judge–repair pipeline plus partial human audit (30% per difficulty).
  • Dual rendering feedback reward (R_visual + R_complex) / RLDRF no independent evidence
    purpose: Internalize aesthetics while preventing oversimplified SVGs during GRPO.
    Paper-specific hybrid reward; evidence is internal ablations and path-tag distribution shifts, not an external law.
  • Generate-Verify-Repair (VRL) agent loop for SVG no independent evidence
    purpose: Supply attribute-level visual feedback and targeted code repair at inference.
    System composition specialized to rendered SVG defects; gains shown via Sankey score transitions on their test distribution.

pith-pipeline@v1.2.0-daily-grok45 · 25893 in / 3871 out tokens · 92552 ms · 2026-07-31T18:40:09.087399+00:00 · methodology

0 comments
read the original abstract

In demanding professional environments and meeting review scenarios, lengthy text often imposes a high cognitive load. To facilitate efficient information communication, transforming verbose text into logically clear diagrams is essential. Scalable Vector Graphics (SVG) provide an effective representation for this purpose due to their editability and resolution independence. However, current research on Text-to-SVG generation remains hindered by three major challenges: (1) the scarcity of datasets for complex, logic-rich diagrams; (2) the absence of explicit layout priors, which leads to chaotic spatial arrangements; and (3) the lack of fine-grained visual feedback to validate rendered outputs and correct aesthetic defects. To address these challenges, at the data level, we introduce DocMeetSVG-100K, a large-scale SVG dataset tailored for document authoring and meeting review scenarios. At the model level, we propose GVR-Coder, a novel framework designed to generate high-quality logical diagrams from lengthy professional texts. Specifically, we adopt a curriculum-driven rejection sampling fine-tuning to progressively enhance the model's capability in modeling complex structures, while explicitly incorporating layout constraint knowledge during training. In addition, we introduce reinforcement learning from dual rendering feedback, a mechanism that provides implicit feedback through reward signals to jointly optimize structural complexity and visual aesthetics. Furthermore, we design a generate-verify-repair agent loop, which improves generation quality through explicit, fine-grained feedback and targeted refinement. Extensive experiments demonstrate that GVR-Coder outperforms competitive baselines and reliably produces logically coherent and visually appealing diagrams. Code and data are available at https://github.com/CurryaNa/GVR-Coder.

Figures

Figures reproduced from arXiv: 2607.28073 by Chunsai Du, Jihua Kang, Qifan Zhang, Qi Song, Tianqi Li, Wangqiu Zhou, Yiming Xu, Yiting Wu.

Figure 1
Figure 1. Figure 1: Motivation. (a) Verbose texts impose high cognitive [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Statistics of our dataset DocMeetSVG-100K. (a) shows the sample count across six primary categories. (b) presents the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The overall framework: Dataset scenarios and GVR-Coder workflow. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Verifier’s Reliability (Left) and Robustness (Right). [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Inference efficiency and performance trade-off. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Factuality Performance Radar Map [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Score transitions across iterative repair rounds. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative performance of GVR-Coder. (Due to space constraints, please refer to the Supplementary Material for [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Prompt used for structured SVG generation. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Prompt used for SVG aesthetic and standardization [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: Prompt used for SVG repair and refinement. [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Effectiveness of dual rendering rewards. [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 16
Figure 16. Figure 16: Quantitative scoring mechanism and multi [PITH_FULL_IMAGE:figures/full_fig_p013_16.png] view at source ↗
Figure 15
Figure 15. Figure 15: Analysis of RLDRF Training Dynamics on n320 datasets. G Rationale for Adopting SVG as the Target Representation We adopt Scalable Vector Graphics as our primary diagram repre￾sentation due to its unique balance of visual expressiveness, precise spatial control, and widespread cross-platform compatibility. While declarative graph languages (e.g., Mermaid) are efficient for standard charts, they are heavily… view at source ↗
Figure 17
Figure 17. Figure 17: Comparative visualization results between Gemini-3-pro and GVR-Coder [PITH_FULL_IMAGE:figures/full_fig_p015_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Comparative visualization results between Gemini-3-pro and GVR-Coder [PITH_FULL_IMAGE:figures/full_fig_p016_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Comparative visualization results between Gemini-3-pro and GVR-Coder [PITH_FULL_IMAGE:figures/full_fig_p017_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Visualization of the Generate-Verify-Repair Agent Loop (VRL) process [PITH_FULL_IMAGE:figures/full_fig_p018_20.png] view at source ↗

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

47 extracted references · 9 linked inside Pith

  1. [1]

    Sumit Asthana, Sagi Hilleli, Pengcheng He, and Aaron Halfaker. 2025. Summaries, Highlights, and Action items: Design, implementation and evaluation of an LLM- powered meeting recap system.Proceedings of the ACM on Human-Computer Interaction9, 2 (2025), 1–29

    2025

  2. [2]

    Alexandre Carlier, Martin Danelljan, Alexandre Alahi, and Radu Timofte. 2020. Deepsvg: A hierarchical generative network for vector graphics animation.Ad- vances in Neural Information Processing Systems33 (2020), 16351–16361

    2020

  3. [3]

    Siqi Chen, Xinyu Dong, Haolei Xu, Xingyu Wu, Fei Tang, Hang Zhang, Yuchen Yan, Linjuan Wu, Wenqi Zhang, Guiyang Hou, et al. 2025. Svgenius: Benchmark- ing llms in svg understanding, editing and generation. InProceedings of the 33rd ACM International Conference on Multimedia. 13289–13296

    2025

  4. [4]

    Zehao Chen and Rong Pan. 2024. SVGBuilder: Component-Based Colored SVG Generation with Text-Guided Autoregressive Transformers.arXiv preprint arXiv:2412.10488(2024)

    arXiv 2024

  5. [5]

    Louis Clouâtre and Marc Demers. 2019. Figr: Few-shot image generation with reptile.arXiv preprint arXiv:1901.02199(2019)

    Pith/arXiv arXiv 2019

  6. [6]

    Google DeepMind. 2025. Gemini 3.0. https://deepmind.google/technologies/ gemini/

    2025

  7. [7]

    Kevin Frans, Lisa Soros, and Olaf Witkowski. 2022. Clipdraw: Exploring text- to-drawing synthesis through language-image encoders.Advances in Neural Information Processing Systems35 (2022), 5207–5218

    2022

  8. [8]

    Adam Hazimeh, Ke Wang, Mark Collier, Gilles Baechler, Efi Kokiopoulou, and Pascal Frossard. 2026. Semantic Document Derendering: SVG Reconstruction via Vision-Language Modeling. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 4636–4644

    2026

  9. [9]

    Qijia He, Xunmei Liu, Hammaad Memon, Ziang Li, Zixian Ma, Jaemin Cho, Jason Ren, Daniel S Weld, and Ranjay Krishna. 2026. VFIG: Vectorizing Complex Figures in SVG with Vision-Language Models.arXiv preprint arXiv:2603.24575(2026)

    arXiv 2026

  10. [10]

    Ajay Jain, Amber Xie, and Pieter Abbeel. 2023. Vectorfusion: Text-to-svg by ab- stracting pixel-based diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 1911–1920

    2023

  11. [11]

    Denis Kocetkov, Raymond Li, Loubna Ben Allal, Jia Li, Chenghao Mou, Car- los Muñoz Ferrandis, Yacine Jernite, Margaret Mitchell, Sean Hughes, Thomas Wolf, et al. 2022. The stack: 3 tb of permissively licensed source code.arXiv preprint arXiv:2211.15533(2022)

    Pith/arXiv arXiv 2022

  12. [12]

    Maikel Leon. 2025. GPT-5 and open-weight large language models: Advances in reasoning, transparency, and control.Information Systems(2025), 102620

    2025

  13. [13]

    Hao Li, Yuhao Wang, Wenning Hao, Pingping Zhang, Dong Wang, and Huchuan Lu. 2026. RAGTrack: Language-aware RGBT Tracking with Retrieval-Augmented Generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 28179–28189

    2026

  14. [14]

    Hao Li, Yuhao Wang, Xiantao Hu, Wenning Hao, Pingping Zhang, Dong Wang, and Huchuan Lu. 2026. Cadtrack: Learning contextual aggregation with de- formable alignment for robust rgbt tracking. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 40. 6109–6117

    2026

  15. [15]

    Jinke Li, Jiarui Yu, Chenxing Wei, Hande Dong, Qiang Lin, Liangjing Yang, Zhicai Wang, and Yanbin Hao. 2025. Unisvg: A unified dataset for vector graphic under- standing and generation with multimodal large language models. InProceedings of the 33rd ACM International Conference on Multimedia. 13156–13163

    2025

  16. [16]

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Cheng- gang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2024. Deepseek-v3 technical report.arXiv preprint arXiv:2412.19437(2024)

    Pith/arXiv arXiv 2024

  17. [17]

    Chaohu Liu, Kun Yin, Haoyu Cao, Xinghua Jiang, Xin Li, Yinsong Liu, Deqiang Jiang, Xing Sun, and Linli Xu. 2024. Hrvda: High-resolution visual document assistant. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 15534–15545

    2024

  18. [18]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. InInternational conference on machine learning. PmLR, 8748–8763

    2021

  19. [19]

    Juan A Rodriguez, Shubham Agarwal, Issam H Laradji, Pau Rodriguez, David Vazquez, Christopher Pal, and Marco Pedersoli. 2023. Starvector: Generating scalable vector graphics code from images.arXiv preprint arXiv:2312.11556(2023)

    Pith/arXiv arXiv 2023

  20. [20]

    Kimi Team, Tongtong Bai, Yifan Bai, Yiping Bao, SH Cai, Yuan Cao, Y Charles, HS Che, Cheng Chen, Guanduo Chen, et al . 2026. Kimi K2. 5: Visual Agentic Intelligence.arXiv preprint arXiv:2602.02276(2026)

    Pith/arXiv arXiv 2026

  21. [21]

    Vikas Thamizharasan, Difan Liu, Matthew Fisher, Nanxuan Zhao, Evangelos Kalogerakis, and Michal Lukac. 2024. Nivel: Neural implicit vector layers for text-to-vector generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4589–4597

    2024

  22. [22]

    Yilin Tian, Yue Shang, Chaoxu Kang, and Kaibo Zhao. 2025. A Review of Text- Guided Vector Graphics Generation. In2025 21st International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD). IEEE, 1–6

    2025

  23. [23]

    Yael Vinker, Ehsan Pajouheshgar, Jessica Y Bo, Roman Christian Bachmann, Amit Haim Bermano, Daniel Cohen-Or, Amir Zamir, and Ariel Shamir. 2022. Clipasso: Semantically-aware object sketching.ACM Transactions on Graphics (TOG)41, 4 (2022), 1–11

    2022

  24. [24]

    Feiyu Wang, Zhiyuan Zhao, Yuandong Liu, Da Zhang, Junyu Gao, Hao Sun, and Xuelong Li. 2025. Svgen: Interpretable vector graphics generation with large language models. InProceedings of the 33rd ACM International Conference on Multimedia. 9608–9617

    2025

  25. [25]

    Gaike Wang, Qiwen Zhao, Zhongwen Zhou, and Yibang Liu. 2025. Research on real-time multilingual transcription and minutes generation for video conferences based on large language models.Spectrum of Research5, 1 (2025)

    2025

  26. [26]

    Haomin Wang, Qi Wei, Qianli Ma, Shengyuan Ding, Jinhui Yin, Kai Chen, and Hongjie Zhang. 2026. Reliable Reasoning in SVG-LLMs via Multi-Task Multi- Reward Reinforcement Learning.arXiv preprint arXiv:2603.16189(2026)

    arXiv 2026

  27. [27]

    Haomin Wang, Jinhui Yin, Qi Wei, Wenguang Zeng, Lixin Gu, Shenglong Ye, Zhangwei Gao, Yaohui Wang, Yanting Zhang, Yuanqi Li, et al. 2025. Internsvg: Towards unified svg tasks with multimodal large language models.arXiv preprint arXiv:2510.11341(2025)

    arXiv 2025

  28. [28]

    Jiuniu Wang, Gongjie Zhang, Quanhao Qian, Junlong Gao, Deli Zhao, and Ran Xu. 2025. Robosvg: A unified framework for interactive svg generation with multi-modal guidance.arXiv preprint arXiv:2510.22684(2025)

    arXiv 2025

  29. [29]

    Yuqing Wang, Yizhi Wang, Longhui Yu, Yuesheng Zhu, and Zhouhui Lian. 2023. Deepvecfont-v2: Exploiting transformers to synthesize vector fonts with higher quality. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 18320–18328

    2023

  30. [30]

    Ronghuan Wu, Wanchao Su, and Jing Liao. 2025. Chat2svg: Vector graphics generation with large language models and image diffusion models. InProceedings of the Computer Vision and Pattern Recognition Conference. 23690–23700

    2025

  31. [31]

    Ronghuan Wu, Wanchao Su, Kede Ma, and Jing Liao. 2023. Iconshop: Text-guided vector icon synthesis with autoregressive transformers.ACM Transactions on Graphics (TOG)42, 6 (2023), 1–14

    2023

  32. [32]

    Ximing Xing, Yandong Guan, Jing Zhang, Dong Xu, and Qian Yu. 2025. Reason- SVG: Hybrid Reward RL for Aha-Moments in Vector Graphics Generation.arXiv preprint arXiv:2505.24499(2025)

    Pith/arXiv arXiv 2025

  33. [34]

    Ximing Xing, Juncheng Hu, Guotao Liang, Jing Zhang, Dong Xu, and Qian Yu

  34. [35]

    Ximing Xing, Juncheng Hu, Jing Zhang, Dong Xu, and Qian Yu. 2024. Svgfu- sion: Scalable text-to-svg generation via vector space diffusion.arXiv preprint arXiv:2412.10437(2024)

    Pith/arXiv arXiv 2024

  35. [36]

    Ximing Xing, Chuang Wang, Haitao Zhou, Jing Zhang, Qian Yu, and Dong Xu

  36. [37]

    Ximing Xing, Haitao Zhou, Chuang Wang, Jing Zhang, Dong Xu, and Qian Yu. 2024. Svgdreamer: Text guided svg generation with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4546–4555

    2024

  37. [38]

    Yiming Xu, Qi Song, Yihan Wang, Wangqiu Zhou, and Junli Liang. 2025. LGC-CR: Few-shot Knowledge Graph Completion via Local Global Contrastive Learn- ing and LLM-Guided Refinement. InProceedings of the 34th ACM International Conference on Information and Knowledge Management. 3688–3697

    2025

  38. [39]

    Yiying Yang, Wei Cheng, Sijin Chen, Xianfang Zeng, Fukun Yin, Jiaxu Zhang, Liao Wang, Gang Yu, Xingjun Ma, and Yu-Gang Jiang. 2025. Omnisvg: A unified scalable vector graphics generation model.arXiv preprint arXiv:2504.06263(2025)

    arXiv 2025

  39. [40]

    Aohan Zeng, Xin Lv, Zhenyu Hou, Zhengxiao Du, Qinkai Zheng, Bin Chen, Da Yin, Chendi Ge, Chengxing Xie, Cunxiang Wang, et al. 2026. GLM-5: from Vibe Coding to Agentic Engineering.arXiv preprint arXiv:2602.15763(2026)

    Pith/arXiv arXiv 2026

  40. [41]

    Peiying Zhang, Nanxuan Zhao, and Jing Liao. 2024. Text-to-vector generation with neural path representation.ACM Transactions on Graphics (TOG)43, 4 (2024), 1–13

    2024

  41. [42]

    Xuanle Zhao, Deyang Jiang, Zhixiong Zeng, Lei Chen, Haibo Qiu, Jing Huang, Yufeng Zhong, Liming Zheng, Yilin Cao, and Lin Ma. 2025. Vincicoder: Unifying multimodal code generation via coarse-to-fine visual reinforcement learning. arXiv preprint arXiv:2511.00391(2025)

    arXiv 2025

  42. [43]

    Yuze Zhao, Jintao Huang, Jinghan Hu, Xingjun Wang, Yunlin Mao, Daoze Zhang, Zeyinzi Jiang, Zhikai Wu, Baole Ai, Ang Wang, et al . 2025. Swift: a scalable lightweight infrastructure for fine-tuning. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 29733–29735

    2025

  43. [44]

    Dawei Zhu, Rui Meng, Yale Song, Xiyu Wei, Sujian Li, Tomas Pfister, and Jinsung Yoon. 2026. PaperBanana: Automating Academic Illustration for AI Scientists. arXiv preprint arXiv:2601.23265(2026)

    arXiv 2026

  44. [45]

    office text-SVG diagram

    Minjun Zhu, Zhen Lin, Yixuan Weng, Panzhong Lu, Qiujie Xie, Yifan Wei, Sifan Liu, Qiyao Sun, and Yue Zhang. 2026. AutoFigure: Generating and Refining Publication-Ready Scientific Illustrations.arXiv preprint arXiv:2602.03828(2026). MM ’26, November 10–14, 2026, Rio de Janeiro, Brazil. Xu et al. Appendix A Overview In this supplementary material, we provid...

    arXiv 2026

  45. [2023]

    Diffsketcher: Text guided vector sketch synthesis through latent diffusion models.Advances in Neural Information Processing Systems36 (2023), 15869– 15889

    2023

  46. [2024]

    arXiv preprint arXiv:2412.11102(2024)

    Empowering LLMs to Understand and Generate Complex Vector Graphics. arXiv preprint arXiv:2412.11102(2024)

    Pith/arXiv arXiv 2024

  47. [2025]

    In Proceedings of the Computer Vision and Pattern Recognition Conference

    Empowering llms to understand and generate complex vector graphics. In Proceedings of the Computer Vision and Pattern Recognition Conference. 19487– 19497