Face_recognition: Classifying large group of people (~100k)

[gp-wang]

I have tried with 500, 1k, and 5k people with the classifier in the example. Small number (500) works with great accuracy, but with large number (5k) the accuracy drops significantly.

Would you give some hint how to organize my classifier for large group of people? (targeting 100k people)

Thank you.

[ofgagliardi]

If you want less false positives, try lowering the "tolerance" argument from the "compare_faces()" function

[EralpB]

I also want to target 100k people, I'm currently fetching 4-5 images per user, let's see how it will work out! Are you trying just similarity matching or like authentication, depending on your usecase, your parameters need to be different!

if someone has best practices, would be glad to hear.

[RajaHaseeb]

Can you share your code of 500 people???

[godcong]

i dead in 1W+ people

[manmitya]

How many people is 1W?

[godcong]

How many people is 1W?

Sorry, I made a mistake..
1W+ means 10K+

[alessiosavi]

Try to hack https://github.com/alessiosavi/PyRecognizer with a lot of photos. It use a Multi Layer Perceptron neural network in order to execute face recognition. I don't know if a large dataset (lot of people with a lot of photos) can lead to confuse the network.