New dataset DisciplineGen-1M boosts AI visual generation for academic content

Researchers have introduced DisciplineGen-1M, a large-scale dataset designed to improve the accuracy of AI models in generating and editing knowledge-intensive visual content. This dataset comprises 1.2 million samples across ten disciplines, including mathematics, physics, and computer science, and was constructed using a framework that combines vector-graphics rendering, OCR-based editing, and programmatic synthesis. Experiments using this dataset have shown significant improvements on discipline-specific benchmarks like GenExam and GRADE, suggesting that structured academic visual data is crucial for advancing AI capabilities beyond aesthetic appeal to verifiable, knowledge-grounded visual creation. AI

IMPACT Enhances AI's ability to generate and edit accurate, knowledge-grounded visual content across academic disciplines.

RANK_REASON The item is an academic paper introducing a new dataset and model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• DisciplineGen-1M
• GenExam
• Gotit.pub
• GRADE
• Hugging Face
• RISE
• ScienceCast
• WISE

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →