Dataset · CASIA-IVA-Lab
VisReason visual reasoning benchmark
CASIA-IVA-Lab/VisReason
A set of 1,505 visual questions to test multimodal model image reasoning capabilities.
What's inside
A benchmark by CASIA-IVA-Lab for evaluating visually-oriented logical inference in everyday scenarios where perception is tightly coupled with reasoning. It contains 1,505 carefully curated image-based questions divided into 10 categories (perceptual, structural, and conceptual reasoning). Created to verify whether multimodal LLMs rely directly on visual cues rather than textual abstractions alone.
What you can do with it
- Evaluating and comparing multimodal language models (VLMs) on Visual Question Answering tasks
- Detecting hallucinations and textual bias during complex scene analysis
- Tasks
- Visual question answering
- Size
- 1K–10K rows
- Downloads this month
- 28.6K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("CASIA-IVA-Lab/VisReason")