Dataset · linxy
Formula images and LaTeX markup
linxy/LaTeX_OCR
Pairs of mathematical formula images with LaTeX code for training and evaluating OCR.
What's inside
Dataset for image-to-LaTeX recognition: formula images and corresponding LaTeX code. The repository has five sets, including small with 110 test examples, full with about 100k printed formulas, and synthetic_handwrite with about 100k synthesized handwritten formulas. Sources are Zenodo, CROHME, and custom data; some LaTeX was filtered out if formulas did not render.
What you can do with it
- Training OCR models for formulas
- Fine-tuning vision-language models on image-to-text
- Testing printed and handwritten formula recognition
- Tasks
- Image → text
- Size
- 100K–1M rows
- Downloads this month
- 909
- Likes
- 195
Models trained on it
- mradermacher/Qwen2-VL-OCR-2B-Instruct-i1-GGUF 1.1K
- mradermacher/Qwen2-VL-OCR-2B-Instruct-GGUF 974
- mradermacher/Qwen2-VL-OCR2-2B-Instruct-GGUF 741
- prithivMLmods/Qwen2-VL-OCR-2B-InstructImage text → text 648
- mradermacher/Callisto-OCR3-2B-Instruct-i1-GGUF 464
- mradermacher/visionOCR-3B-061125-GGUF 289
- mradermacher/Callisto-OCR3-2B-Instruct-GGUF 245
- prithivMLmods/visionOCR-3B-061125Image text → text 57
How to load it
from datasets import load_dataset
ds = load_dataset("linxy/LaTeX_OCR")