Chemical Reaction Images from Scientific Literature

10,512 chemical reaction images were extracted from scientific literature PDFs. The dataset, created by songjhPKU, is designed for chemical reaction diagram parsing to support cheminformatics and computer vision research.

Use Cases

• Train a computer vision model to parse chemical reaction diagrams from 10,512 images.
• Develop document analysis tools for extracting structured chemical data from scientific PDFs.
• Build a model for automatic captioning or annotation of chemical reaction images.

Strengths

• Contains 10,512 annotated chemical reaction images.
• Provides a dedicated split with 10,112 images for training and 400 for testing.
• Specifically designed for parsing chemical diagrams, a niche application in cheminformatics.

Limitations

• Limited test set size of 400 images may not be sufficient for robust model evaluation.
• Unknown annotation schema and column structure limits understanding of data granularity.
• Potential bias towards reaction types and literature sources from which PDFs were extracted.

Provenance

Source

songjhPKU on Hugging Face.

Collection Method

Images extracted from scientific literature PDFs.

Freshness

Last updated on 2025-10-31.