Skip to content

Dataset · echo840

Benchmark for evaluating OCR quality in multimodal models

echo840/OCRBench

Image and text sets for testing models' text recognition capabilities.

What's inside

The dataset is a popular benchmark for evaluating optical character recognition (OCR), with a paper accepted in Science China Information Sciences. It contains image-text pairs, and the dataset size ranges from 1k to 10k examples.

What you can do with it

  • Test text recognition accuracy in VLMs and multimodal LLMs.
  • Compare various OCR systems on a unified reference set.
Size
1K–10K rows
Downloads this month
27K
Likes
26

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("echo840/OCRBench")