Skip to content

Dataset · antins-labs

VisSearch-Bench Visual Search Benchmark

antins-labs/VisSearch-Bench-v1

204 English questions with 190 images for evaluating long visual search in multimodal agents.

What's inside

The English test portion of the VisBench-EN benchmark consists of 204 questions divided into five task categories, and 190 input images (including multi-image sets for group search). The structure, identifiers, and order of images are fully mapped to the Chinese version VisBench-CN.

What you can do with it

  • Evaluation of multimodal agents in complex multi-step image search tasks
  • Cross-language comparison of model visual reasoning quality in English and Chinese
Tasks
Image text → text
Languages
English
Size
up to 1K rows
Downloads this month
1.2K
Likes
7

How to load it

from datasets import load_dataset

ds = load_dataset("antins-labs/VisSearch-Bench-v1")