Poldion Self-HostedDeutsch
Part of Poldion Self-Hosted

FaceRec

Local face recognition for the doorbell in Home Assistant: "Anna is at the door." No cloud, learning via a button in the notification.

View source Quick start
  • Python
  • FastAPI
  • ONNX Runtime
  • InsightFace
  • Home Assistant
  • AMD ROCm
FaceRec UI with fictional people

The screenshot shows the German interface – the app itself can be switched to English.

How it works

When someone rings, Home Assistant sends the picture to the service. It finds the faces, compares them with the known people and answers with a sentence for a push notification or speech output. Unknown faces can be assigned to a person straight from the notification; similar unknown faces are grouped.

For a hallway camera the service can read a video stream for a few seconds and track every face. That way each person is reported once, even if several arrive at the same time.

Features

Fast

~90 ms per image on the CPU, a few milliseconds with an AMD GPU.

Keeps learning

Assign unknown faces, train from photos, rename and merge people.

Correctable

"Wrong person" via a notification button. Recognised faces are kept for 48 hours for that.

Home Assistant

Ready-made package with REST commands and sensors, templates for the automations.

Privacy: face data is biometric data. Run the service only in your own network, tell visitors about the camera and follow the applicable data protection laws (in the EU the GDPR).

Working together

FaceRec is a standalone service for Home Assistant. If it runs on the same machine as the OllamaProxy, it attaches there as a module, with a link in the UI and its memory in the view. It has nothing to do with the tabletop part of the family.

language model (LLM) prompts · module LLM LLM recaps tokens, maps, scenes characters & lore module OllamaProxy language models ImageGen images & videos WhisperNotes transcription BasaltNotes knowledge base Tabletop Forge game table FaceRec face recognition
OllamaProxyAttaches FaceRec as a moduleImageGenImages & videos with ComfyUIWhisperNotesTranscription & session recapsBasaltNotesNotes & world wiki with an AI curatorTabletop ForgeDigital game table for 5e

Quick start

git clone https://github.com/Poldion/FaceRec.git && cd FaceRec
mkdir -p ~/facerec/data && cp facerec.env.example ~/facerec/facerec.env   # set HA_URL and HA_TOKEN
docker compose --env-file ~/facerec/facerec.env up -d --build
# with an AMD GPU: add -f docker-compose.rocm.yml

Then open http://<host>:8191/ and add the package homeassistant/facerec.yaml to Home Assistant.

License

MIT license for the code. The InsightFace recognition models (buffalo_l) are licensed for non-commercial research only and are not included.