Benchmarking scientific figures as editable powerpoint slides
ReFigBench: Benchmarking Scientific Figure Reconstruction as Editable PowerPoint Artifacts
Computation and LanguageComputer Vision and Pattern Recognition
Summary
Turning a complex scientific image into an editable PowerPoint slide is a tricky task because the slide must keep the text, layout, and connections intact. The authors looked at how well different AI coding agents can recreate these figures as editable slides through two methods. They tested 1,000 real scientific figures and found that even the best AI struggles to perfectly balance how accurate and editable the recreated slides are. Their work shows that both the AI model and the software environment around it impact the quality of the final slide.
What this means in practice
- •For presentation designers: Generate editable scientific slides automatically from images to speed up creation and updates of presentations.
- •For document automation teams: Integrate AI agents that reconstruct figures as editable PowerPoint objects, improving document reuse and editing.
- •For visual content software developers: Develop tools that enable seamless figure reconstruction inside presentation software using specialized AI workflows.$Commercial implications: Enables development of software products that automate figure conversion into editable slides for commercial office suites.
Authors
Liyang Fan, Chi Wei, Yitai Li, Xinping Bi, Guhong Chen, Chenghao Sun, Haoxiang Yang, Qingwen Li, Kai Yan, Hong Li, Bo Li
Abstract
Multimodal coding agents are expected to turn visual inputs into usable artifacts, and they act through a harness, the layer of tools, context management, and execution environment around the model. Existing evaluations often isolate short tool calls, API traces, or screenshot resemblance, and a low score under these proxies cannot say whether the model saw poorly, planned poorly, or was failed by its harness. We study scientific overview figure reconstruction, an agent task in which a source image must become an editable PowerPoint slide that preserves text, topology, layout, and native document structure. We introduce ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance. Coding agents from four model families reconstruct every figure under two workflows, direct code generation and a specialized PPTX workflow, and the strongest model runs inside two commercial harnesses, yielding ten configurations. Evaluation combines deterministic artifact checks, repeated automated scoring by judges from two model families, and blinded human comparisons. Perception remains a bottleneck that iterative rendering only partly repays. Whether workflow effort converts into quality depends on the model together with its harness, since the same model gains from the specialized workflow inside one harness and loses inside the other, and the harness shifts scores even under an identical direct prompt. The specialized workflow erases native connectors in every configuration, human judges still prefer its renderings in most matchups, and even the strongest agent falls short of the rubric ceiling. These results expose the tension between fidelity and editability as the central challenge for practical multimodal document agents.