Match 1:N about deepface HOT 10 CLOSED

1- store registration pictures in a folder (e.g. c:\users\dan\desktop\db).

Suppose that db folder has 3 identities: 1.jpg, 2.jpg, 3.jpg

2- when an user comes your system, take his/her current photo. Suppose that the current photo is tmp.jpg

3- build a for loop with the number of identities in your db folder. Call DeepFace for each photo in your db.

for i in range(0, 3):
   obj = DeepFace.verify("tmp.jpg", "%s.jpg" % (str(i)))
   if obj["verified"] == True:
      print("This is identity", i)
      break

I might add a find function to handle this operation in a simpler way.

from deepface.

Thanks for your quick reply. : D

Okay, I can loop it, so how much is it expensive for the CPU? Taking into account that the user does not want to wait too long.

Another question, taking advantage of the same context: is there a way to pre-train or pre-process (I don't know exactly the term) so that when making this loop I pass not the image, but a string?
For example, see this project https://github.com/ageitgey/face_recognition the images are nothing but an array of face information.

P.S .: I don't know exactly the terms because I'm trying to find a solution, I don't know if recognition is for authentication and identification, I'm lost.

from deepface.

Nope. all pre-process steps handled in the background of deepface.

The code block I previously shared builds a complex face recognition model in each iteration. This is costly. On the other hand, prediction is easy.

If you call deepface as illustrated below, then it will build a complex face recognition model once. This speeds your operations up dramatically.

obj = DeepFace.verify([
["tmp.jpg", "1.jpg"],
["tmp.jpg", "2.jpg"],
["tmp.jpg", "3.jpg"]
])

Then, obj variable is a list. You should check its verified and distance values.

from deepface.

Okay, I tried here.

I got the result:

{'pair_1': {'distance': 0.1356278657913208,
            'max_threshold_to_verify': 0.4,
            'model': 'VGG-Face',
            'similarity_metric': 'cosine',
            'verified': True}}

Which of these keys do I use to tell if it's the same person or not?

from deepface.

verified. If it is true, then the both images are same person.

from deepface.

Thank you for your time and congratulations on the project. I will test everything you showed me.

from deepface.

you're welcome. do not hesitate to ask questions if you have any problem.

PS: you can support this project by starring the repo :)

from deepface.

I star the project now :)

If your project fits mine, I'll pay you a lot of beers haha Hugs

from deepface.

As I mentioned, I can modify the framework based on the requirements. Inform me if you need anything.

from deepface.

I have the same question above, I'm building an attendance system but it takes an intensive amount of time for matching each image with the images in my dataset. Is there a way where I can speed up the process? maybe by saving the images as numpy and pass it to verify?

Thank you

from deepface.