New pipeline unlocks visual data in materials science literature

Researchers have developed MatMMExtract, an open-source pipeline designed to unlock the visual data within materials science literature. This system decomposes complex scientific figures into individual sub-panels and generates structured annotations using a large language model and a specialized taxonomy. Applied to over 14,000 articles, it created MatSciFig, a dataset of nearly 400,000 image-text pairs, each with detailed categorization and summaries. The project also introduced MaterialScope, a detection dataset that improved a YOLO12-m model's accuracy for localizing figure panels, and found Gemini 3.1 Flash Lite to be the most cost-effective LLM for generating annotations. AI

IMPACT Enables large-scale AI analysis of visual data in scientific literature, potentially accelerating materials science discovery.

RANK_REASON The cluster describes a new dataset and pipeline for processing scientific literature, fitting the research category. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Gemini 3.1 Flash Lite
• Hugging Face
• MaterialScope
• MatMMExtract
• MatSciFig
• YOLO12-m

• paper
• product
• infra

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