Researchers have developed an open-source pipeline called MatMMExtract that extracts image-text pairs from materials science literature. The pipeline decomposes complex scientific figures into individual sub-panels and generates structured annotations using a large language model. Applied to 14,810 articles, it produced over 391,000 panel-level image-text pairs. A related dataset, MaterialScope, was also introduced for accurate panel localization. The work aims to make the visual record of materials science knowledge more accessible to AI. --- Why it matters: This matters because it enables researchers and engineers to access a large-scale, structured dataset of image-text pairs from materials science literature, which can be used to improve vision-language learning and downstream applications in areas like materials discovery and design. Source: https://arxiv.org/abs/2606.29667