Image-to-molecule benchmarking dataset with fractal pattern and hierarchical morphology recognition.

The unique phenomenon of high morphological diversity of quaternary phosphonium salts (QPSs) has been observed via electron and optical microscopy. The molecular structure of the QPSs, which differ by one methylene group, was shown to be reflected in the microstructure of the crystallized droplets. Here, we describe experimental datasets of scanning electron and optical microscopy images at different magnifications, illustrating the versatile microstructures of 19 homologous QPSs. The unique patterns that appear in the microscopy images of the QPS are related to the molecular structure. The described datasets of microscopy images are made openly available for scientific purpose and include hierarchical morphological patterns and fractal elements. Importantly, the datasets are suitable for both directions of machine learning exploration: recognizing molecular formulas from microscopy images and, conversely, predicting morphological patterns from molecular structures. This bidirectionality establishes a benchmark for bridging the molecule-morphology gap and advancing data-driven materials design.