swiss-ai/handwriting_data

Handwriting LaTeX Dataset

Paired PNG images and LaTeX source for handwriting recognition.

Dataset Structure

• data split: 49,882 samples from handwritten notes on paper backgrounds
• data_slides split: 5,100 samples from beamer/slide-style presentations

Each sample contains:

• image: A PNG rendering of handwritten LaTeX content
• latex: The corresponding LaTeX source code
• image_filename: The PNG filename encoding variant info (text color, page type, blur, noise, grid)
• content_id: Numeric ID linking image variants to their shared LaTeX source

Usage

from datasets import load_dataset

ds = load_dataset("username/handwriting_data", trust_remote_code=True)

sample = ds["data"][0]
sample["image"]          # PIL Image
sample["latex"]          # LaTeX source string
sample["image_filename"] # e.g. "overlayed_merged_content_1_blacktext_ML4Science_paperpage_grid.png"
sample["content_id"]     # e.g. 1