Skip to content

Dataset · mixed-modality-search

Multimodal Text and Image Search Benchmark

mixed-modality-search/MixBench25

Set of queries and mixed corpora based on MSCOCO, Google_WIT, VisualNews, and OVEN for cross-modal search evaluation.

What's inside

The benchmark combines data from four well-known sources: MSCOCO, Google_WIT, VisualNews, and OVEN. It contains English texts and images divided into splits with queries, corpora (corpus, mixed_corpus), and relevance assessments (qrel).

What you can do with it

  • Evaluating the quality of mixed multimodal information retrieval systems (text-ranking)
  • Testing embeddings working simultaneously with text and visual documents
Tasks
Text ranking
Languages
English
Downloads this month
48.7K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("mixed-modality-search/MixBench25")