face_recognition

One-Shot Learning and Face Recognition (Implementation of FaceNet Inception model)

Siamese Network for One-Shot Learning :

A Siamese network is an architecture with two parallel neural networks, each taking a different input, and whose outputs are combined to provide some prediction

Triplet Loss function for Learning Face Embeddings:

The loss function penalizes the model such that the distance between the matching examples is reduced and the distance between the non-matching examples is increased.

Weights:

Keras-OpenFace pretrained weights are used for minimizing triplet loss function

Euclidean distance:

Calculates distance between the test image vector with all the images in the database. If distance is less than the threshold factor, then prediction is correct

New-user:

Implemented a function to start web-cam and capture users faces, crop them for training