mol-image-to-smiles

To convert chemical structure images to SMILES strings, you input molecular images in formats like PNG or JPG to extract molecule recognition data. The system processes single or batch images and outputs JSON files containing SMILES and model confidence.

Molecule recognition supports PNG, JPG, JPEG, GIF, and BMP image formats. You can process these formats individually or in batches to generate SMILES strings with confidence metrics.

DECIMER operates locally as the primary model, while MolNextR via HuggingFace serves as a fallback. This dual-model approach provides fault tolerance and allows comparing different prediction results for batch images.

Yes, you can batch process multiple chemical images and export the results. The system outputs JSON files containing SMILES strings and model confidence information to a configurable directory.

Yes, the dependencies include RDKit and huggingface_hub. DECIMER is the primary local model, but HuggingFace is required for the MolNextR fallback mechanism when local processing fails.