Skip to content

Dataset · moonshotai

Visual World Knowledge Benchmark for MLLMs

moonshotai/WorldVQA

Image-based questions to evaluate atomic world knowledge in multimodal models.

What's inside

WorldVQA is a benchmark for evaluating atomic visually-centric world knowledge in multimodal LLMs. It decouples visual knowledge retrieval from reasoning to precisely measure what a model knows. It includes images and questions in English and Chinese, with a size ranging from 1k to 10k examples.

What you can do with it

  • Evaluating MLLMs on visual world knowledge
  • Comparing vision-language models
  • Diagnosing the separation of retrieval and reasoning
Tasks
Visual question answering
Languages
English, Chinese
Size
1K–10K rows
Downloads this month
2.3K
Likes
67

How to load it

from datasets import load_dataset

ds = load_dataset("moonshotai/WorldVQA")