aiqbench analysis
This research addresses the critical gap in iterative refinement for scientific figures by leveraging naturally occurring revision trajectories from arXiv and GitHub to train specialized editing models. By utilizing a large-scale dataset of TikZ edits and reinforcement learning, the authors demonstrate that compact Qwen3.5-based models can achieve performance rivaling or exceeding current proprietary VLMs on publication-ready tasks. The work establishes a new benchmark and open-source resources specifically for the niche but essential domain of scientific figure generation and modification.
EdiTikZ: Scientific Figure Editing from Revision Trajectories
arXiv:2609.01409v1 Announce Type: new Abstract: Vision-language models (VLMs) have shown strong performance in generating scientific figures from text or images. However, producing publication-ready figures requires iterative refinement, making scientific figure editing an important yet largely unexplored task. Existing approaches rely on costly proprietary agentic systems, focus primarily on evaluation, or construct training supervision from synthetically generated edits. Instead, we leverage naturally occurring scientific revision and development trajectories as a scalable source of supervision. To this end, we introduce DaEdiTikZ, the first large-scale dataset of revision-derived scientific figure edits, constructed by mining 391K plausible TikZ edit pairs from arXiv, GitHub, and TeX SE and inferring 781K directed edit instructions with a VLM conditioned on rendered figures and TikZ code. We further introduce DaEdiTikZ-Bench, a human-refined benchmark with 790 instances, and train two compact Qwen3.5-based EdiTikZ models (4B and 9B) by jointly learning reconstruction and editing, followed by reinforcement learning (RL) with complementary rewards for rendered fidelity and edit application. Automatic evaluation places our 9B model above all tested baselines, while human evaluation with 9 annotators and 4,320 ratings places it above GPT-5.6-Sol and on par with Gemini-3.1-Pro. Under severe out-of-distribution shifts, it remains competitive with GPT-5.6-Sol near its 2K training sequence-length regime. Models and datasets will be released.
Abstract:Vision-language models (VLMs) have shown strong performance in generating scientific figures from text or images. However, producing publication-ready figures requires iterative refinement, making scientific figure editing an important yet largely unexplored task. Existing approaches rely on costly proprietary agentic systems, focus primarily on evaluation, or construct training supervision from synthetically generated edits. Instead, we leverage naturally occurring scientific revision and development trajectories as a scalable source of supervision. To this end, we introduce DaEdiTikZ, the first large-scale dataset of revision-derived scientific figure edits, constructed by mining 391K plausible TikZ edit pairs from arXiv, GitHub, and TeX SE and inferring 781K directed edit instructions with a VLM conditioned on rendered figures and TikZ code. We further introduce DaEdiTikZ-Bench, a human-refined benchmark with 790 instances, and train two compact Qwen3.5-based EdiTikZ models (4B and 9B) by jointly learning reconstruction and editing, followed by reinforcement learning (RL) with complementary rewards for rendered fidelity and edit application. Automatic evaluation places our 9B model above all tested baselines, while human evaluation with 9 annotators and 4,320 ratings places it above GPT-5.6-Sol and on par with Gemini-3.1-Pro. Under severe out-of-distribution shifts, it remains competitive with GPT-5.6-Sol near its 2K training sequence-length regime. Models and datasets will be released.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2609.01409 [cs.AI]
(or arXiv:2609.01409v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.01409 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 306 words
Related coverage
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · Jun 12, 2026
GenGA: Editable and Data-Grounded Graphical Abstract Generation for Academic Papers
arXiv · Aug 7, 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · Aug 26, 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · Oct 13, 2026