SciForma: Structure-Faithful Generation of Scientific Diagrams
2026-07-20 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionGraphicsMachine Learning
AI summaryⓘ
The authors focus on improving the accuracy of scientific methodology diagrams, which must correctly show components, arrows, and text to make sense. They identify that existing models often fail because they don’t ensure all parts are right at the same time. To fix this, they created SciForma, a system that evaluates and trains models using three key parts of a diagram and introduces a new method (M-DPO) to fix errors across all parts simultaneously. This approach helps SciForma produce more reliable scientific diagrams than previous open-source models and some proprietary ones. They also provide datasets and tools for others to build on their work.
structural fidelityscientific methodology diagramssupervised fine-tuningMulti-Dimensional Conjunctive Preference Optimizationdiagram componentsdirectional relationstextual annotationsiterative editinglogic-verified evaluationopen-source models
Authors
Yuxuan Luo, Peng Zhang, Xinjie Zhang, Xun Guo, Zhouhui Lian, Yan Lu
Abstract
Structural fidelity is essential to scientific methodology diagrams. To communicate research logic, these diagrams must faithfully render components, directional relations, and textual annotations. Since a single error, such as a reversed arrow or an unreadable equation, can invalidate the entire figure, structural fidelity is inherently conjunctive: correctness on one axis cannot compensate for failure on another. Current open-source models fail to satisfy this criterion. Supervised fine-tuning (SFT) learns plausible layouts but cannot reliably ensure structural correctness, while scalar reward-based post-training obscures which structural dimension has failed. To address this, we introduce SciForma, a framework for the structure faithful generation of scientific methodology diagrams. Specifically, SciForma decomposes diagram quality into three structural axes: Component, Arrow, and Text, guided by a structural inventory. Built on this foundation, we curate SciFormaData-700K for structured training and SciFormaBench-2K for logic-verified evaluation. To close the gap left by SFT, we develop Multi-Dimensional Conjunctive Preference Optimization (M-DPO), which enforces simultaneous correctness across all axes and adaptively routes gradients to the most deficient dimension in post-training. The same structural inventory also enables iterative editing at inference time to correct residual errors. This combination allows SciForma-9B to exceed all open-source baselines and GPT-Image-1.5 on both SciFormaBench-2K and AIBench, bringing open scientific diagram generation close to proprietary-level structural fidelity. Our code and data will be available at: https://github.com/microsoft/SciForma.