AI BriefingAI That Understands Chemical Molecular Structure Information in Documents, MolMole Unveiled - LG AI Research Blog
Key point
LG AI Research has unveiled MolMole, a specialized model that directly extracts chemical molecular structures and reaction formulas from PDF documents.
Details
LG AI Research has introduced Deep Document Understanding (DDU) technology that understands not only text, graphs, and tables within documents but also complex chemical molecular structural formulas and reaction formulas. As its latest achievement, it unveiled MolMole, a model specialized for the chemistry field.
Unlike existing models, MolMole reflects real-world research needs by allowing the PDF document itself to be used as input, without the need to separately crop images. This enables researchers to extract chemical data from complex patents or papers much more efficiently.
MolMole consists of three core modules:
- ViDetect: Precisely detects molecular structure regions within a document
- ViReact: Recognizes chemical reactions and the composition of molecular structures
- ViMore: (details omitted)
Additionally, LG has built its own benchmark dataset that, going beyond existing single-model performance evaluation (OCSR), can measure the ability to extract molecular structures from entire PDF documents in a manner similar to real chemical research environments, and plans to release it later this year.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.