Wikipedia LaTeX Formulas: 319K Pairs of Code and Rendered Images

An October 2025 snapshot of English Wikipedia yielded approximately 319,000 mathematical formulas. This curated collection provides LaTeX source code paired with high-quality rendered images, created by author piushorn and last updated on Hugging Face in March 2026. Formulas were filtered by visual complexity and renderability.

Use Cases

• Training formula recognition models based on the paired LaTeX code and rendered images.
• Benchmarking OCR systems for mathematical content based on the high-quality formula images.
• Developing tools for automatic extraction and rendering of mathematical notation from text sources.
• Creating synthetic training data for document understanding models based on the structured formula pairs.

Strengths

• Contains approximately 319,000 data points, providing a substantial corpus.
• Pairs LaTeX source code with corresponding high-quality rendered images, offering structured multimodal data.
• Formulas are filtered by a visual complexity score (>8), suggesting a focus on non-trivial examples.
• Sourced from a complete extraction of English Wikipedia, providing a broad base of mathematical notation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is an approximation (~319k), and the exact data scale may require verification.
• The dataset's snapshot is from October 2025; freshness for specific applications should be verified.

Provenance

Source

English Wikipedia articles.

Collection Method

Complete extraction and curation, filtered by visual complexity and renderability.

Time Range

Snapshot from October 2025.

Freshness

Last updated 2026-03-27 10:34:04; source snapshot from October 2025.