Dataset on which I worked on 5 Celebrity Faces Dataset then added some of my photos to expand the dataset. Then I used 'haar cascade classifier' for face detection and finally trained the data by adding 2 custom layers on top of MobileNetV2 CNN
The Loss function used is binary cross-entropy is uses sigmoind fuction to classify the data. it predicts for each the probabily of it being in postive class and negetive class. The accuracy in validation data is ~86% with batch size of 32
Install Jupyter - https://jupyter.org/install (Or Open in google collab https://colab.research.google.com)
The .ipynb file has the code. Download the dataset unzip it, also download the 'haarcascade_frontalface_default.xml' in the root folder.
Numpy Pandas Tensorflow Keras OpenCV 2 Matplotlib