In this notebook about face recognition, a pipeline is covered from preprocessing
images to recognizing the faces they contain.
In the process, images are 'embedded' in a vector space. By doing so, we can cluster images of an individual together with low in-between distances,
while keeping a further distance relative to the images of another.
The embedding into a 128-dimensional vector space is done by a convolutional network called FaceNet.
After reducing these dimensions from 128 to 3 with Pincipal Component Analysis, it is possible to visualize them.
The result is the visualization below.
Here you can add your own embeddings.
Your webcam will shoot a number of frames, calculate the embedding per frame, as well as an ideal embedding approximation.
Make sure to turn your head for different angles!
Here you can upload an image to verify or recognize.
Face of Obama Recognized.