How do we use different FacialRecognition models e.g., Evision.Face.LBPHFaceRecognizer

[faizann]

I am trying to replicate this blog post in Python which uses LBPHFaceRecognizer.
In Python code once you create an instance of LBPHFaceRecognizer you get to call train function on it.
There is no such function in Elixir bindings in LBPHFaceRecognizer. Does this mean that it has to be manually coded or have I missed something?

[cocoa-xu]

Hi @faizann, the module you're looking for in Elixir is [Evision.Face.LBPHFaceRecognizer](https://hexdocs.pm/evision/Evision.Face.LBPHFaceRecognizer.html). If you're using smart cell in Livebook, then it's not there yet. However, you can follow the similar approach as the Evision.FaceRecognizerSF

[faizann]

Hi @cocoa-xu

Thanks for your quick response. I am not using SmartCell rather just elixir app or livebook.
I have looked at Evision.FaceRecognizerSF where features are taken and then can be matched.

The Evision.Face.LBPHFaceRecognizer does not have feature or match or train methods.

This is from my livebook

face_recogniser = Evision.Face.EigenFaceRecognizer.create()
Evision.Face.FaceRecognizer.train(face_recogniser, [mat],0)

Output

{:error,
 "cannot get `Ptr<cv::face::FaceRecognizer>` from `self`: mismatched type or invalid resource?"}

I feel like I am missing something obvious but couldn't figure it out after spending a few hours on the problem.

[cocoa-xu]

Thanks for the error message! I probably know how to address this particular error.

{:error,
 "cannot get `Ptr<cv::face::FaceRecognizer>` from `self`: mismatched type or invalid resource?"}

I'll try to fix it today or tomorrow and release a new version

[cocoa-xu]

Hi @faizann, evision v0.2.8 is shipped and the following code should work :)

images = 
  [
    "person0/0.jpg", "person0/1.jpg", "person0/2.jpg",
    "person1/0.jpg", "person1/1.jpg", "person1/2.jpg"
  ]
  |> Enum.map(&Evision.imread(&1, flags: Evision.Constant.cv_IMREAD_GRAYSCALE()))
  |> Enum.map(&Evision.Mat.as_type(&1, :s32))

labels = Nx.tensor([0, 0, 0, 1, 1, 1], type: :s32)

face_recogniser = Evision.Face.EigenFaceRecognizer.create()
Evision.Face.EigenFaceRecognizer.train(face_recogniser, images, labels)

[faizann]

That is awesome. Thanks a lot for fixing it.