ChemVA: Advancing Large Language Models on Chemical Reaction Diagrams Understanding

In three linesChemVA framework advances LLM understanding of chemical reaction diagrams through Visual Anchor mechanism for functional group detection and semantic alignment translating visual features to entity names. Achieves 92.0% structural recognition accuracy on OCRD-Bench dataset and ~20 percentage point performance gain across 9 diverse LLMs.

Summary generated by Claude — human-verified