Research map / Large language models / Retrieval and knowledge
Diagram2Structure: Unlocking LLMs' Diagram Comprehension through DiagramDiff, a Framework for Structuring Offline Diagrams
Haoxiang Hu, Yaokun Li, Zeyuan Huang, Cangjun Gao, Qiang He, Qingkun Li, Xiaoming Deng, Cuixia Ma, Yu-Kun Lai, Yong-Jin Liu, Hongan Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add diffusion-based feature enhancement (FE) to the graph transformer recognizer (ablation) | DiagramDiff recognition · CASIA-OHFC (reconstructed strokes) | SCA (%): 90.56 -> 94.64 | Improved |
| add diffusion-based feature enhancement (FE) (ablation) | DiagramDiff recognition · OHSD (reconstructed strokes) | SCA (%): 94.33 -> 98.39 | Improved |
| recognize strokes reconstructed from images instead of original online strokes (ablation) | SpaceGTN · CASIA-OHFC | SCA (%): 98.13 -> 89.56 | Hurt |
| recognize reconstructed instead of original strokes (ablation) | InstGNN · OHSD | SCA (%): 96.89 -> 91.24 | Hurt |
| use proposed stroke reconstruction instead of prior reconstruction methods | DiagramDiff reconstruction · CASIA-OHFC | IoU (%): 50.5 -> 55.6 | Improved |
| use diffusion-GTN recognizer instead of SpaceGTN | stroke recognition · CASIA-OHFC (reconstructed strokes) | SCA (%): 89.56 -> 94.64 | Improved |
| feed DiagramDiff structured diagram data to GPT-4o alongside the image | GPT-4o · DiagramQAE editing | accuracy (%): 71% -> 90% | Improved |
| feed DiagramDiff structured data to Claude 3.7 | Claude 3.7 · DiagramQAE Q&A | accuracy (%): 54% -> 68.5% | Improved |
| feed DiagramDiff structured data to DeepSeek R1 | DeepSeek R1 · DiagramQAE Q&A | accuracy (%): 69% -> 80% | Improved |
| feed DiagramDiff structured data to GPT 4.5 | GPT 4.5 · DiagramQAE Q&A | accuracy (%): 77.5% -> 93% | Improved |
Closest papers by meaning
- Math Blind: Failures in Diagram Understanding Undermine Reasoning in MLLMs (ICLR 2026 · similarity 0.86)
- CoG-DQA: Chain-of-Guiding Learning with Large Language Models for Diagram Question Answering (CVPR 2024 · similarity 0.85)
- Do Vision-Language Models Really Understand Visual Language? (ICML 2025 · similarity 0.85)
- From Words to Structured Visuals: A Benchmark and Framework for Text-to-Diagram Generation and Editing (CVPR 2025 · similarity 0.83)
- VisioMath: Benchmarking Figure-based Mathematical Reasoning in LMMs (ICLR 2026 · similarity 0.83)
- ChartGalaxy: A Dataset for Infographic Chart Understanding and Generation (ICLR 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
