Database

LanceDB-based storage for face embeddings and attributes.

What is ReID Database?

ReID_Database is the underlying storage layer for face embeddings and person attributes in degirum-face. It uses LanceDB to store:

• Face embedding vectors for similarity search

• Person attributes (names, metadata)

• Associated face images

• Model metadata for validation

The database is automatically managed by FaceRecognizer and FaceTracker - you typically don't instantiate it directly.

When to Use

Automatic usage - The database is created and managed when you initialize FaceRecognizer or FaceTracker:

# Database created automatically at db_path
recognizer = degirum_face.FaceRecognizer(
    config=degirum_face.FaceRecognizerConfig(
        db_path="./my_faces.lance",  # Database location
        # ...
    )
)

Direct usage - Only needed for:

• Custom database operations

• Database migration/backup tools

• Advanced batch processing

• Direct embedding management

How It Works

The database uses LanceDB with four internal tables:

Table	Purpose
embeddings	Face embedding vectors with image references
attributes	Person metadata (names, IDs)
images	Face image storage (PNG format)
_metadata	Model validation metadata

Key features:

• Vector search - Fast similarity search using cosine distance

• Deduplication - Automatic embedding deduplication via hashing

• Thread-safe - Safe for concurrent access

• Model validation - Ensures embeddings from same model

• Connection pooling - Single connection per database path

See Configuration Guide for database initialization options.

See Methods Reference for complete API.