Face Shape Detection

[rbp15]

Didn't see a place to ask a question - so using this issue type.

Can this solution detect a person's face shape?

Square, Triangle, Oval, Heart, etc.?

[pospielov]

Hi, unfortunately, no. Do you know any examples of applications that may need it?

[rbp15]

Thank you for replying!

The situation is where I would use the solution for eyeglasses try on.
Some eyeglass providers have face detection worked into there "try-on" solution where the face shape is detected and then glasses frames that go with that shape are recommended.

Example about glasses and face shape: https://www.zennioptical.com/glasses-for-face-shape https://www.foreyes.com/media/blog/face-shapes.png https://www.clearly.ca/thelook/glasses-for-face-shape/

[rbp15]

@pospielov Thoughts? I see you're on AWS Marketplace, that would be a great add there. Rekognition can't do it right now.

[pospielov]

Looks interesting. I'll add a story for this. But now sure if this is an easy task. The main problem is that it should be ready to use a neural network model that we can integrate into CompreFace. I hope we will be able to find it.

[rbp15]

I was looking into:

• https://docs.opencv.org/3.4/da/d60/tutorial_face_main.html
• https://rajathithanrajasekar.medium.com/opencv-series-4-dlib-68-point-face-landmark-prediction-172cb867c869
• https://github.com/PacktPublishing/Python-Deep-Learning-Projects/tree/master/Chapter10
• https://github.com/rajendra7406/FaceShape/blob/master/faceshape.py
• https://www.kaggle.com/search?q=shape_predictor_68_face_landmarks.dat

Also thank you!

[pospielov]

In all of those links, they just find face landmarks. And this is what we already have in CompreFace. Furthermore, I don't think landmarks can help with defining face shape, because the algorithm assumes every face has the same shape when it finds the landmarks.