Facial Recognition Using Resnet50

Facial recognition is done by genrating facial embeddings and then minimizing the distance between the facial embeddings the Concept of Siamese Networks is used to train the model. The loss fuction used is Binary Cross Entropy

Siamese Network - https://medium.com/wicds/face-recognition-using-siamese-networks-84d6f2e54ea4

Resnet50 - https://towardsdatascience.com/the-annotated-resnet-50-a6c536034758

Prerequisites

It is advisable to create a separate python enviorment to install all packages to avoid conflict

pip install opencv-python
pip install tensorflow
pip install path
pip install torch
pip install scikit-learn

Steps for Facial Recognition

Step 1 - Facial Detection

we will be using MTCNN for facial detetion to install mtcnn use the command

 pip install mtcnn 

( Make sure you have the compatible version of opencv and tensorflow installed before intalling mtcnn (if not mtcn will install opencv itself))

MTCNN - https://towardsdatascience.com/face-detection-using-mtcnn-a-guide-for-face-extraction-with-a-focus-on-speed-c6d59f82d49

Step 2 - Facial Alignment

MTCNN can perform facial alignment

Step 3 - facial recognition

Training

The model is trained on the LFW dataset ( make sure that you copy all the images in each of the folder - negative ,positive and anchor )

In this model no pretrained weights are used and is made from scratch

Dataset

If you want to train on you whole dataset just remove .take() in the data pipline this will ensure the model takes input of the whole data. Though it is recomended only if you have GPU

Prediction

Finaly the model can predict the faces of people in real time