Skip to content

Dataset · CSU-JPG

Millions of text-bearing images for image generation

CSU-JPG/TextAtlas5M

A set of 1 to 10 million images and English text annotations for training text-to-image models in text rendering.

What's inside

The dataset includes 1 to 10 million pairs of reference images and English text descriptions collected to train the TextAtlas model. It features various data categories: synthetic stylized and clean text, presentations, book covers, research papers, long words, and scenes with text.

What you can do with it

  • Training text-to-image models to correctly render text on images
  • Fine-tuning LoRA for generating posters, covers, and slides
  • Training text recognition and analysis systems in complex visual scenes
Tasks
Text → image
Languages
English
Size
1M–10M rows
Downloads this month
26.1K
Likes
40

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("CSU-JPG/TextAtlas5M")