Skip to content

Dataset · vikhyatk

LNQA image question answering

vikhyatk/lnqa

A VQA dataset based on Localized Narratives featuring image question-answer pairs.

What's inside

A visual question answering (VQA) dataset built on Localized Narratives. It contains images and text question-answer pairs with a volume ranging from 100k to 1M examples. Details of the compilation are described in the author's blog.

What you can do with it

  • Fine-tuning VQA models such as LLaVA variants
  • Evaluating multimodal models on image-based questions
  • LoRA adaptation of vision-language models
Tasks
Visual question answering
Size
100K–1M rows
Downloads this month
813
Likes
91

How to load it

from datasets import load_dataset

ds = load_dataset("vikhyatk/lnqa")