0.9.1b1

Better

• Counters accept the value after a restart or update. How much a counter may go up now grows
  with the time it was not seen — the app was away, or the readings were unsure — so the first
  reading after an update is no longer rejected as changed too much. A wheel read one digit too
  high over and over still stays out.
• A counter learns its own limit (Learn it, Settings → Sensor output): how fast it goes up
  on a busy day, learned from a week of its readings. Or set a fixed rise per hour. Counters
  you already have keep the limit you set, turned into a rate.
• A real jump is accepted once several readings agree (the Live tab says Jump accepted) —
  say the light failed during a shower. A jump that looks like one misread digit waits for you in
  Review.
• Read correctly in Review moves the counter on: answering it for the newest rejected reading
  makes that the value — also the way out for a counter stuck on a value read too high.
• Switching the face model shows its progress on the sensors' cards, the Faces card, People and
  Settings, and the taught faces are learned again right after the new model is loaded.

0.9.0

New

• Recognise people by their face on object sensors that look for Person. Off until you turn it
  on per sensor; the faces never leave your Home Assistant.
  • Name the people of your household by tapping them on the Live tab, in the history or in the
    review queue, or add photos from your phone on the new People page.
  • A person named on one camera is known on every camera, with their own sensor that says where
    they are now. Unknown person turns on for a clear face that matches nobody.
  • While someone walks up, the sensor looks for their face about every second, and when it is
    not sure who it is, it asks you in Review: each answer teaches it how that camera sees them.
  • Made for doorbells: faces in front of a bright sky are evened out before they are compared,
    and the Live tab shows whether your camera sees faces well enough.
  • Two face models: SFace (fine on a Raspberry Pi 4) and AuraFace (for a fast PC).
• Larger AI models for fast PCs under Settings → AI model: D-FINE M and L find more
  small and far-off objects, DINOv2 base tells similar states apart best. Compare every model's
  memory, speed and accuracy there, and delete downloaded ones you no longer use.
• Teaching limits you can tune, for all object sensors and per sensor.
• The wizard warns about a small picture, which may be the camera's low-resolution stream.

Better

• The History shows what you taught from it (Taught: Our car).
• "Last seen" survives an update or restart: a sensor no longer says Not seen yet about
  something it has seen before.

Fixed

• Boxes on the Live tab now always match the picture shown.
• Switching AI models gives the memory of the old model back without a restart.

0.9.0b5

Fixed

• Opening the People page could fail right after a face was taught (an error about
  face_embedding), when a check was comparing faces at the same moment.

0.9.0b4

Better

• Each person has a "last seen" sensor. Next to Ole seen by (which camera sees them now; was
  just Ole), Ole last seen shows in Home Assistant when they were last seen, also when you
  confirmed afterwards that a frame was them. New installs name them
  sensor.visionstate_ole_seen_by and sensor.visionstate_ole_last_seen; entities you already
  have keep their IDs.
• Answering a question marks the whole moment in the History: the face seen entry of the
  same frame says Taught: Ole too.

0.9.0b3

Better

• Faces are looked for while someone is there. Someone walking up to a doorbell shows a usable
  face for only a few seconds, often between two checks. When a check finds a person nobody is
  named for, a sensor that recognises faces now checks about every second while they come closer,
  until they are named or gone (half a minute at most, and once per visit).
• The History shows what you taught from it. After naming a person or correcting a box in a
  history frame (or answering a question in Review), its entry says Taught: Ole.
• "Last seen" survives an update or restart. A sensor no longer says Not seen yet about Our
  car or a person it has seen before: when each was last seen comes back from the history.

0.9.0b2

Better

• Faces in front of a bright sky are recognised better. A doorbell looking out at the sky
  often sees a face that is nearly black; its light is now evened out before it is compared, so
  you are named sooner and other people look less alike.
• It asks more when it is not sure who someone is, so answering teaches it how each camera
  sees you: from a lower similarity (with SFace), several times in a row while someone walks up
  (a few seconds apart, up to five), and right away again after you answer.
• After importing a sensor that recognised faces, the app says that face recognition is off
  on the new sensor and opens its Settings tab to turn it on.
• The text of Ask me about faces I do not know says what it does.

0.9.0b1

Fixed

• A quiet log when faces are recognised. Each time the SFace face model was loaded, the app's
  log filled with 174 harmless warnings from ONNX Runtime about how the model file is stored. The
  model works exactly as before; only errors from ONNX Runtime are logged now.

0.8.1b5

Fixed

• Boxes on the Live tab now always match the picture. While a check was still looking for
  objects and faces, the Live tab could show its new frame with the boxes of the check before — a
  person (and their face) drawn where nobody stood any more. A new frame now appears together with
  what was found in it.
• The face model is let go when face recognition is turned off while a check was still busy:
  that check could load it again, and it then stayed in memory until a restart.

0.8.1b4

New

• Teach a face from your phone. On the People page, Add a person or Add photos on a
  person takes photos straight from the phone's camera or its gallery (also in the Home Assistant
  app), a short video turning the head a little, or a ZIP of photos. Each goes through the same
  check as a face on a camera, and the message says what was added and why a photo was skipped.
  A quick start; faces taught on each camera still recognise people best.

0.8.1b3

New

• Recognise people by their face on object sensors that look for Person. Off until you turn it
  on, per sensor, after a sheet that says what is stored and what is downloaded (once, from Hugging
  Face); it can also be turned on in the new sensor wizard. The faces never leave your Home
  Assistant.
  • Name the people of your household by tapping them on the Live tab, in a history frame or in
    the review queue: Who is this? Only a usable face can be taught, so a back or a jacket never
    becomes someone.
  • Shared by every camera: a person named on one camera is known on all that recognise faces.
    Each gets an on/off sensor and a count per camera, and their own sensor
    (sensor.visionstate_<name>) that says which camera sees them now, or not_seen.
  • Unknown person turns on for a face seen clearly that matches nobody you named; a face
    further away or turned to the side can still be named, but never makes someone a stranger.
  • People (in the menu) lists everyone with their faces and the camera each was taught on,
    where they were last seen, people the AI finds alike and faces that may carry the wrong name;
    rename, move or forget faces and people there.
  • Does my camera see faces well enough? The Live tab draws each face and says why one cannot
    be used (too small, turned away, cut off by the edge, blurry), and how many of the last people
    had a usable face.
  • A face nobody is named for keeps a frame in the history (Face seen, not named), so someone
    who was only at the door for a moment can be named afterwards.
  • Limits for faces (smallest face, how close a face must be to be named, asked about or kept)
    under Settings → Recognising faces for all sensors, and per sensor.
  • Two face models: SFace (small, fine on a Raspberry Pi 4) and AuraFace (for a fast PC).
• Compare and manage the models (Settings → AI model): every model from the smallest to the
  largest, with the face models next to the object detector; a downloaded model that is not in use
  can be deleted there.
• The wizard warns about a small picture: it may be the camera's low-resolution stream, while
  the main stream is often hidden in Home Assistant.