Skip to content

Dataset · facebook

TextVQA: Text-based Visual Question Answering

facebook/textvqa

Facebook benchmark of 45 thousand image questions to test neural network reading and text comprehension in frames.

What's inside

It contains 45,336 English questions for 28,408 images taken from the open OpenImages dataset. The task requires the model not just to recognize text in a photo, but to comprehend it to provide the correct answer.

What you can do with it

  • Evaluation of multimodal models (VLM) in visual-question-answering tasks focusing on OCR
  • Fine-tuning models to recognize and interpret inscriptions on signs, labels, and documents
Tasks
Visual question answering
Languages
English
Size
10K–100K rows
Downloads this month
1.3K
Likes
38

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/textvqa")