ITMol-500K molecular image-text dataset

ITMol-500K is a large-scale molecular image-text dataset constructed for molecular property prediction and cross-modal retrieval. It contains approximately 500,000 molecular image-text pairs. Molecular images were generated from molecular structures, and textual descriptions were derived from molecular annotations and generated captions. This dataset was used in the study “ITMol: A Molecular Image-Text Foundation Model Bridging the Semantic Gap for Property Prediction and Retrieval”.