Double Take

Double Take is a unified web UI and REST API for processing, reviewing, and training images used for facial recognition. It abstracts differences between multiple face detection/recognition backends and provides a single workflow for matches, unknowns, and subject training.

Key Features

• Web UI for reviewing matches and managing training/untraining images per subject
• REST API for submitting images and retrieving stored match/train/latest images
• Works with multiple detectors (for example CompreFace, DeepStack, CodeProject.AI Server, Amazon Rekognition, Facebox)
• NVR integration support (notably Frigate) to process snapshots and events
• MQTT subscribe/publish for automation workflows and Home Assistant discovery
• Optional authentication to protect the UI and API, with access tokens
• Scheduling controls to disable detection during defined time windows
• Optional image preprocessing using OpenCV

Use Cases

• Add facial recognition review and training to Frigate-based home video surveillance
• Publish recognition results over MQTT to trigger Home Assistant automations and notifications
• Provide a single API façade when switching or testing different face recognition backends

Limitations and Considerations

• Requires an external detector service; Double Take orchestrates and manages workflows rather than performing recognition entirely on its own
• Recognition accuracy and performance depend heavily on the chosen detector and camera image quality

Double Take is best suited for users who want a consistent UI and automation-friendly API for facial recognition workflows, especially when integrating NVR events and MQTT-based home automation. It simplifies multi-backend setups and centralizes training and match management in one place.