I'm relatively new in machine learning and I am trying to put together my undergraduate thesis on masked face recognition.
I've read Ejaz and Islam's paper (available at https://www.researchgate.net/publication/340690545_Masked_Face_Recognition_Using_Convolutional_Neural_Network) about feature extraction using FaceNet and Facial Verification using SVM, and I am stuck on understanding some parts of the paper.
What is a very deep CNN network supported by L2 normalization and how does it work?
Is triplet loss the only way to increase the accuracy of a CNN architecture?
How do someone retrain the FaceNet model? Isn't a model already trained and can recognize something out of the box?
How does the SVM work in facial verification?
What do they mean by "A masked face similarity is measured upon the masked and non-masked face by estimating an L2 normalization within the features key points collected from the net structure"?
In short, I am confused about the method of their paper as they don't really elaborate much about it.
Thanks in advance guys, I really appreciate some help because I'm really stuck on it.
