This notebook demonstrates the development of a siamese neural network used for facial identification.

It was trained with 10 datasets (LFW, FERET, AT&T...) and tested with 1 dataset (MUCT), resulting in more than 700,000 images (350,000 pairs).

Saving all the imagens in a .h5 file and sharing in this repository was not possible, because most of these datasets don't allow redistributions.

Each model of the siamese neural network has 8 convolutional layers. Dropout was also used to avoid overfitting.

The result for this network was 84% accuracy on the MUCT dataset.

The direct visualization of some of the results (distances between faces) can be seen in the end of the notebook.