Using Viseron 3.0.0b8 with CodeProject.AI but unable to train as no images get saved to the unknown folder...

[sgoldtho]

Using Viseron 3.0.0b8 with CodeProject.AI but unable to train as no images get saved to the unknown folder.

Folder structure: face_recognition/faces/unknown

Face recognition config for codeprojectai:

face_recognition:
  save_unknown_faces: true
  expire_after: 10
  cameras:
    front_porch:
  labels:
    - person

The DEBUG logs for vireson.components.codeprojectai.face_recognition.front_porch show...

[2024-07-07 15:10:19] [DEBUG ] [viseron.components.codeprojectai.face_recognition.front_porch] - Face recognition result: {'message': 'No known faces', 'count': 1, 'predictions': [{'confidence': 0, 'userid': 'unknown', 'x_min': 64, 'y_min': 19, 'x_max': 116, 'y_max': 108}], 'success': True, 'inferenceMs': 129, 'processMs': 131, 'moduleId': 'FaceProcessing', 'moduleName': 'Face Processing', 'code': 200, 'command': 'recognize', 'requestId': '13f42a44-c558-416a-9c4a-c91a31e19cd6', 'inferenceDevice': 'CPU', 'analysisRoundTripMs': 135, 'processedBy': 'localhost', 'timestampUTC': 'Sun, 07 Jul 2024 05:10:19 GMT'}

Have I missed something?

[roflcoopter]

They should be stored along with the other detections, by default that is in /snapshots/face_recognition/<CAMERA IDENTIFIER>/unknown

[sgoldtho]

Ah ha, at least I can see my camera has a folder under face_recognition, but no subfolders under the camera.

I might have some 'serious' configuration issues as I went from version 2 to version 3 of Viseron using the same config file and Docker settings.

Have the Docker Recording Path and Config Path changed between the two versions?

Update... Re-configure my docker run to align with what you have here: https://dev--viseron.netlify.app/docs/documentation/installation#running-viseron

Will see what happens...

[sgoldtho]

With the corrected docker settings I can see images being saved in /snapshots/face_recognition/front_porch/unknown and can move them to thier repective subfolders.

Some observations:

If I restart Veseron to get the faces to train the default seems to be '/config/face_recognition/faces/' so I added 'face_recognition_path: /snapshots/face_recognition' but that returned 'No images were found for face front_porch in folder /snapshots/face_recognition/front_porch. Please provide some images of this person'.

I changed to 'face_recognition_path: /snapshots/face_recognition/front_porch' and then the training could find the images in the sub-folders.

codeprojectai
  face_recognition:
    face_recognition_path: /snapshots/face_recognition/front_porch
    save_unknown_faces: true
    expire_after: 10
    cameras:
      front_porch:
    labels:
      - person

Question:

Is that how it is supposed to be configured or is there something not quite right?

[roflcoopter]

You should not have to change face_recognition_path. Did you use face recognition before v3? The only thing that has changed is where detected faces are stored (in snapshots) so it should not affect your training data

[sgoldtho]

Yes I have used face recognition before v3.

I'll remove face_recognition_path from my config then tomorrow will double check where things go...

First thing I've noticed is that without face_recognition_path being set Viseron reads from '/config/face_recognition/faces/' not from '/snapshots/face_recognition/'

[sgoldtho]

I've remapped '/config' to a new directory on my host system and restarted Viseron and everything seems to be working now (I think).

Images get saved to snapshots/face_recognition/<camera name>/unknown for unrecognised, or snapshots/face_recognition/<camera name>/<person name> for recognised.

Then for the training, copy images to config/face_recognition/faces/<person name> and restart Viseron to pickup any new images.

[roflcoopter]

Yes that is how it is intended to work! In the future i hope to make it managable through the frontend instead of having to manually copy files.

Gonna close this if you dont mind, feel free to reopen if you encounter issues