Skip to content

Dataset · UniParser

Nearly 8M pairs of molecule images and SMILES formulas

UniParser/MolParser-7M

Large corpus of nearly 8 million molecular structure images annotated in extended SMILES format for model training and evaluation.

What's inside

The dataset contains nearly 8 million image-text pairs, mapping chemical molecule images to their extended SMILES descriptions. The main training split includes over 7.7 million synthetic examples, with evaluation data also provided. Materials prepared by the paper's authors for ICCV 2025.

What you can do with it

  • Training multimodal models for visual recognition of chemical structures (molecular OCR)
  • Fine-tuning VLM models to translate formula images into machine-readable SMILES format
  • Testing the parsing accuracy of molecular diagrams
Size
1M–10M rows
Downloads this month
2.1K
Likes
33

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("UniParser/MolParser-7M")