Task Family

Multimodal Retrieval

Canonical ID https://id.searchplex.net/multimodal-retrieval/

Retrieval involving information needs, indexed items, representations, or results that span more than one data modality, such as text, images, audio, video, tables, or structured metadata.

Also called

  • Multimodal Search
  • Multimedia Retrieval
  • Multimedia Information Retrieval

Scope

Multimodal retrieval is a broad family. It includes settings where multiple modalities are searched together, represented jointly, or used in query and result interpretation. It is distinct from cross-modal retrieval, where the query and target results are in different modalities, and from image retrieval, where the retrieved objects are images.

Observed terms

  • Multimodal Vector Search vendor
  • Neural Multimodal Search industry

Part of family

Includes

Not equivalent to

  • Multimodal Retrieval is not Cross-Modal Retrieval. Multimodal retrieval is the broader family involving multiple modalities; cross-modal retrieval specifically crosses a modality boundary between query and retrieved items.
  • Multimodal Retrieval is not Image Retrieval. Image retrieval targets images; multimodal retrieval is broader and may involve images, text, video, audio, tables, or combinations of modalities.
  • Multimodal Retrieval is not Multi-Vector Retrieval. Multimodal retrieval is about data modalities; multi-vector retrieval is about representing a query or item with multiple vectors.
  • Multimodal Retrieval is not Retrieval-Augmented Generation. Multimodal retrieval may be used inside multimodal RAG systems, but RAG is a retrieval-augmented generation pattern.
  • Multimodal Retrieval is not TRECVID. TRECVID is an evaluation campaign for video analysis and retrieval; multimodal retrieval is a broader retrieval task family.

Learn more

ID