Chemical Information Extraction Benchmark from Patent Images

MolMole_Patent300 is a curated benchmark introduced in the 'MolMole: Molecule Mining from Scientific Literature' paper for evaluating models that extract chemical information from full-page patent images. It enables end-to-end evaluation covering molecule bounding box detection, reaction parsing, and optical chemical structure recognition (OCSR).

Use Cases

• Train and evaluate object detection models for molecule bounding box detection in patent images.
• Benchmark optical chemical structure recognition (OCSR) systems on full-page patent document images.
• Develop and test models for reaction parsing from scientific literature and patent figures.

Strengths

• Designed as a benchmark for end-to-end evaluation of chemical information extraction from images.
• Covers three specific tasks: molecule detection, reaction parsing, and optical chemical structure recognition.

Limitations

• The dataset size, number of rows, and specific file formats are unknown, limiting assessment of scale.
• Sample data and column structure are unavailable, making initial data exploration difficult.

Provenance

Source

Introduced in the paper 'MolMole: Molecule Mining from Scientific Literature' by author doxa-friend.

Collection Method

Curated benchmark dataset, likely compiled from full-page patent images for model evaluation.

Freshness

Last updated on July 21, 2025.