DisciplineGen-1M: A Large-Scale Dataset for Multidisciplinary Visual Generation and Editing

2026-07-02

AI Summary

Existing image generation and editing models lack reliability when handling knowledge-intensive diagrams that require domain-specific expertise, symbolic structures, and precise spatial relationships. To address this limitation, this work introduces the first large-scale, structured multimodal visual dataset comprising 1.2 million samples across ten academic disciplines—including mathematics, physics, and chemistry—and proposes a scalable data generation framework that integrates vector graphic rendering, procedural synthesis, OCR-driven editing, and large-scale filtering. Building upon this dataset, the authors develop a generative model augmented with domain-aware reasoning, significantly enhancing the factual fidelity of generated content. Experiments demonstrate substantial improvements over open-source baselines on the GenExam and GRADE domain-specific benchmarks, while also showing strong transfer performance on the general reasoning benchmarks WISE and RISE, thereby underscoring the critical role of structured academic visual data in knowledge-driven image generation.

Abstract

Recent image generation and editing models can produce visually appealing natural images, yet they remain unreliable when the target image is a knowledge-intensive diagram whose correctness depends on disciplinary concepts, symbolic structure, and precise spatial relations. We introduce DisciplineGen-1M, a million-scale multidisciplinary dataset that supports text-to-image generation and image editing. It contains 1.2M samples spanning mathematics, physics, chemistry, biology, geography, computer science, economics, history, music, and sports. To construct the dataset, we design a scalable framework that combines vector-graphics rendering, OCR-based editing, curated programmatic synthesis, and large-scale text-to-image filtering. These pipelines produce captions, editing instructions, structured annotations, and paired images with controllable semantic differences. Building on DisciplineGen-1M, we further introduce a discipline-informed reasoning-generation model for both text-to-image generation and image editing. Experiments on discipline-related benchmarks, GenExam and GRADE, show substantial improvements over open-source baselines, while evaluations on general reasoning-informed benchmarks, WISE and RISE, further indicate broader transfer. The results suggest that large-scale structured academic visual data is a key ingredient for moving image generation from aesthetic plausibility toward verifiable knowledge-grounded visual creation. We will publicly release our dataset, model, and source code of the data curation pipeline to ensure reproducibility and benefit future research.

Problem

• knowledge-intensive diagrams
• discipline concepts
• symbolic structure
• spatial relations
• visual generation

Innovation

• multidisciplinary visual generation
• structured academic dataset
• knowledge-grounded image editing
• discipline-informed reasoning
• scalable data curation