Skip to content

Dataset · ChongyanChen

Real StackExchange visual question-answering dataset

ChongyanChen/VQAonline

64 thousand real image-question-answer pairs from StackExchange for VQA tasks with verified answers.

What's inside

The VQAonline dataset includes 64 thousand visual questions with images sourced from real discussions on StackExchange communities. Unlike synthetic datasets, each example features natural clarifying context and an answer confirmed by the question author.

What you can do with it

  • Training and validating Vision-Language Models (VLMs) on real queries
  • Evaluating model capability to account for textual context during image analysis
Tasks
Visual question answering
Size
10K–100K rows
Downloads this month
208.5K
Likes
19

How to load it

from datasets import load_dataset

ds = load_dataset("ChongyanChen/VQAonline")