Facial Verification using a Siamese Neural Network

Hello! This is the facial verification/recognition repository you came across. I used one-shot learning using a siamese neural network to prevent training a model, which can be time consuming and consume resources.

Model Architecture

The neural network itself is a pretty simple one. Here's the architecture:

• ConvLayer - 64 3x3 filters with strides of 2
• MaxPool2D layer - 2x2 "filter"
• Dropout with 20% Probability
• ConvLayer - 128 2x2 filters (no padding/strides)
• MaxPool2D layer - 2x2 "filter"
• Flatten Layer
• Fully Connected Layer with 128 logistic regression units (layer that outputs encodings)