Skip to main content

What is ApertureDB?

ApertureDB stores images, videos, documents, feature vectors, and their metadata. Its query language combines property filters, graph relationships, and vector similarity search. Queries can also process images and videos during ingestion or retrieval.

Accelerating AI Development with ApertureDB

ApertureDB Core Architecture​

png

Four components provide storage, search, and data processing:

  • Vector Database: Feature vectors, also called descriptors or embeddings, represent images, text, documents, or video frames numerically. ApertureDB searches these vectors by similarity, with indexing support built on FAISS.
  • Graph Database: Objects, their properties, and their connections represent application metadata. This graph connects media, embeddings, and annotations to the information needed to search them. Bounding boxes and other regions of interest associate labels with parts of an image.
  • Multimodal Data Management: Images and videos support operations such as cropping, sampling, and thumbnail generation through libraries including OpenCV and ffmpeg. Documents, audio, and text can be stored as Blobs without media-specific processing operations. Data can reside on disks available to the server or in cloud object stores.
  • Unified Query Interface: A JSON query combines commands across these components. The query engine dispatches the commands and assembles their results and binary outputs into a response.

Applications use this interface to add, modify, and retrieve data, construct datasets, and supply data to training, inference, and RAG workflows. Distributed deployments add components for horizontal scaling and high availability.

Key Features​

  • Graph transactions: Graph mutations within a query share a transaction. Account-management changes, response construction, and transport have separate failure boundaries, described in the response reference.
  • Storage options: The data-management layer reads and writes cloud object storage, including S3 and GCS, or POSIX-compliant file systems.
  • Data augmentation: Queries can transform images and videos on retrieval, reducing the need to store copies in every required format. Video operations also provide access to individual frames.
  • Combined search: The graph schema represents media, metadata, and relationships. Descriptor sets provide indexing methods and distance metrics for similarity search and classification. Nearest-neighbor queries can also apply metadata constraints.
  • Annotations, labels, and captions: Bounding boxes and polygons identify regions of interest in images or frames. Region operations can return just the associated pixels. Clips and frames identify parts of a video without requiring separate copies, and properties attach captions or other labels to stored objects.
  • Visualization: The ApertureDB UI displays datasets and supports visual inspection of query results.
  • AI/ML integrations: Batch access supplies data to training and inference pipelines. Integrations connect ApertureDB with tools for ingestion, processing, and generative AI workflows.