Handwritten Mathematical Expressions for LaTeX Conversion

A collection of handwritten mathematical expression images paired with LaTeX code, created by Azu and hosted on Hugging Face. The dataset contains between 10,000 and 100,000 samples, as indicated by its size category, and was last updated in March 2022. It is designed for training models to convert visual mathematical notation into structured text markup.

Use Cases

• Train an image-to-text model to generate LaTeX code from handwritten expression images.
• Develop a symbol segmentation model using the image data to isolate individual mathematical operators and digits.
• Benchmark the accuracy of LaTeX sequence generation against the ground-truth text labels.
• Create a multimodal classifier to identify common mathematical expression types from the image and text pairs.

Strengths

• Contains between 10,000 and 100,000 samples based on the '10 Kn100 K' size category tag.
• Provides a direct mapping between two modalities: image data and corresponding LaTeX text.
• Hosted on a major platform (Hugging Face) with a structured metadata record (Croissant).

Limitations

• The exact number of rows, column names, and data distribution are unknown.
• Data may be geographically biased, as indicated by the 'Regionus' tag suggesting United States origin.
• The dataset has not been updated since March 2022, potentially limiting model relevance for newer benchmarks.

Provenance

Source

Hugging Face, uploaded by author Azu.

Collection Method

Presumably collected as images of handwritten expressions with manually or synthetically generated LaTeX labels.

Time Range

null

Freshness

Last updated in March 2022; update frequency is unknown.

Geography

Likely United States, inferred from the 'Regionus' platform tag.