MolGallery: 10 Million Chemical Structure Images from Real-World Documents

10 million chemical structure images sourced from real-world documents like patents, scientific literature, books, and websites. MolGallery captures complex noise, diverse rendering styles, and scanning artifacts absent in synthetic datasets. The dataset was created by UniParser and last updated on March 17, 2026.

Use Cases

• Training OCSR models based on real-world chemical structure images
• Benchmarking model robustness based on diverse rendering styles and scanning artifacts
• Developing image preprocessing techniques based on complex noise patterns from scanned documents
• Analyzing chemical structure representation based on sources like patents and scientific literature

Strengths

• 10 million images provides a large-scale training resource
• Images are sourced from real-world documents, capturing authentic noise and artifacts
• Diverse sources include patents, scientific literature, books, and websites

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Data may reflect source bias inherent to the specific documents collected

Provenance

Source

UniParser

Collection Method

Sourced from real-world documents including patents, scientific literature, books, and websites.

Time Range

null

Freshness

Last updated 2026-03-17 12:05:04; freshness should be verified

Geography

null