VDMSL

VDMSL is a specialized query and control language designed for the Visual Data Management System (VDMS). It provides a formal interface to store, retrieve, and manipulate multimedia data and associated metadata within VDMS, enabling users to describe datasets, annotations, and processing tasks in a consistent way. Through VDMSL, applications can express operations such as adding media objects, linking metadata, configuring feature extraction, and issuing searches and aggregations over large collections.

VDMSL supports a layered data model that includes media items (images and video frames), features, annotations, and provenance data. It accommodates constraints across content, time, spatial relations, and metadata, allowing complex queries such as finding items with specific labels within a time window or retrieving items by feature similarity. The language is designed to support both declarative queries and procedural orchestration of processing steps, making it suitable for end-to-end AI data workflows.

Architecturally, VDMSL statements are interpreted by a VDMS server. Clients interact with the server through official libraries in languages such as Python and C++, which translate VDMSL requests into lower-level operations on storage backends and metadata stores. This arrangement helps separate data modeling and query expression from the underlying data management primitives, enabling scalable performance and reproducible workflows.

Development and usage: The VDMS project aims to provide scalable data management for AI workloads, and VDMSL is its dedicated interface for defining and executing data-centric tasks. It is used in research and production settings to organize large media repositories, enable reproducible experiments, and streamline data-driven workflows. The project is maintained by a community of researchers and practitioners.