Real-Time face recognition on CPU with Python

In this video, I'll introduce a simple way to build and use any custom face recognition model with my custom framework. After completing this tutorial, you will learn how to use pre-trained models to create a real-time face recognition system with any CPU.

Officially, Face Recognition is defined as the problem of verifying or identifying faces in an image. But usually, there is a question, how exactly do we recognize a human face in an image?

There are many ways to implement each step in a face recognition pipeline. In this article, we will focus on a popular deep learning technique for face detection using the MediaPipe face detection model covered in the first tutorial, feature extraction using FaceNet, and face matching with cosine or euclidean distance.

#facerecognition #opencv #machinelearning

[bouachaisraa]

Hi, The code works correctly but it does not recognize me

[sakshamverma]

Hey there great video. I have been trying to this project but the issue im facing is converting that model to onxx. If possible can you please state what version of tensorflow and tf2onxx do i require to make it work on ubuntu. Thanks in advance

[happysunvn]

I tested it and there were quite a few errors. Can you please give me information about specific versions of opencv-python
mediapipe
stow
numpy
tqdm
onnxruntime
tf2onnx
Is tensorflow okay? Thanks a lot

[yosayaan]

how to get the faceNet.onnx file on models folder??

after download and move the facenet_keras_weights.h5 file in models folder. . .
please reply

[Nico-hfhw]

This is fantastic! Just a question, why is it that when I provide multiple photos per person the accuracy goes extremely low? Whilst with one photo per person is really accurate

[nicolassabumathew]

Greetings Sir!, Thank you for the tutorial but could you please provide the facenet.onnx file required ? I am unable to execute the code due to not being able to convert and i have tried a lot of methods.
Thanks
*EDIT*
The issue got fixed, i downgraded tf2onnx to version 1.4.1 and it worked

[Nico-hfhw]

Is it possible to train more than one photo per face?
Like if I wanted to add like 10 photos of one person to make it more accurate

[cando_jo]

Hi! Why can't you just give us the .onnx file? I am getting so much errors using the tf2onnx I think it's outdated or it doesnt work well with the last module in Tensorflow