search

esimov / pigo

Fast face detection, pupil/eyes localization and facial landmark points detection library in pure Go.
MIT License
4.39k stars 310 forks source link

Question: Facial/Identity Recogntition possible? #23

Closed prologic closed 5 years ago

prologic commented 5 years ago

Would it be possible to extend this library or use it as-is to perform facial recognition and thus identity of persons in images? (as opposed to face detection)

esimov commented 5 years ago

No, it's not possible. The library does not support facial recognition.

prologic commented 5 years ago

Thanks I admit I don't know enough about this area to know what it takes to supprot this. Would this be outside of the scope of pigo entirely?

esimov commented 5 years ago

Currently yes, because Pigo is using pretrained data (cascades) for face, eyes and facial landmark points detection. In this sense Pigo is much like a computer vision than a machine learning library.

At the moment i'm not considering extending the library in that direction, but who knows maybe in the future.

olekukonko commented 4 years ago

Now that pigo supports Facial landmark points detection, Is it not possible to save the face’s attributes and vector and then use Euclidean distance to implement facial recognition?

olekukonko commented 4 years ago

@esimov what do you think ?

esimov commented 4 years ago

Interesting point of view, but from my opinion the difficulty of this approach is that the face recognition outcome always have to be independent of way the person is displayed regardless of image dimension, the face angle, the mouth position etc. which means I don't foreseen this approach to provide viable results without involving neural networks.

olekukonko commented 4 years ago

@esimov it might interest you that this approach has been used before with the meaningful result however i agree with you 100% that neural networks is the way to go.

esimov commented 4 years ago

Thanks for the link. Dlib by default supports face recognition and as the article states they are using Dlib on their application. But anyway thanks for coming up with this idea, it's worth to make some research on this topic. Any PR's, feedback, articles are welcome.

olekukonko commented 4 years ago

@esimov a pure go facial recognition is definitely gold. I would definitely keep discussing this with you. Well done for the excellent work.

olekukonko commented 4 years ago

Any update on this?