Skip to content

Dataset · Salesforce

BLIP3 multimodal OCR-focused dataset

Salesforce/blip3-ocr-200m

Large-scale dataset of images with text and descriptions from Salesforce for training VLM models.

What's inside

The dataset containing between 10 and 100 million examples includes image-text pairs rich in text (documents, charts) and English text data. Optical character recognition (OCR) methods were integrated during creation to accurately convey details.

What you can do with it

  • Pretraining vision-language models (VLMs) for document and chart understanding
  • Improving the quality of text recognition and analysis in images (OCR)
Languages
English
Size
10M–100M rows
Downloads this month
3.9K
Likes
47

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Salesforce/blip3-ocr-200m")