search

dcmartin / motion-ai

AI assisted motion detection for Home Assistant
http://motion-ai.com/
Apache License 2.0
282 stars 29 forks source link

Close feedback loop #16

Open dcmartin opened 3 years ago

dcmartin commented 3 years ago

  1. Close feedback loop from front-end (end-user interaction) to annotation capture and curation platform. a. Enable labelstud.io as additional Home Assistant add-on [X] b. Enable min.io as additional Home Assistant add-on [X] c. Enable end-user feedback through HA persistent_notification with link to yes/no, including all relevant metadata [X] d. Capture feedback from MQTT topic, store object in minio and store metadata in labelstudio add-ons

  2. Present captured feedback for end-user interaction a. review & approve b. inspect, create, update, delete putative annotations

  3. Present training results a. review & approve b. inspect, delete training results

  4. Deploy updated trained models a. Add modelmanager service to control OH edge-sync-service b. Add hzn::ess and hzn::modmgr SDK components to base containers c. Modify AI containers to utilize hzn::modmgr SDK for model weights, labels, etc..

dcmartin commented 3 years ago
  1. Completed end-user interaction for Home Assistant to notify through standard persistent notifications.
  2. Added functionality to Motion Classic add-on to provide API to respond to predict.yes and predict.no which bundle up jpeg, mp4, and metadata for non-legacy events and post to MQTT as <group>/predict/{yes,no}
dcmartin commented 3 years ago
  1. Implemented labelstudio add-on for Home Assistant as OH service base with bashio extensions; amd64
  2. Implemented minio add-on for HA as OH service using base with bashio extensions; amd64 and aarch64