Complete dataset can be found here : https://fex.insa-lyon.fr/get?k=DMBBtY0Z90zvqMtRqOZ (expired the 12/12/2019)

This dataset countains both training and test dataset for a total of 163 100 pictures (78 800 face and 84 300 non-face) in 36x36 format.

In order to work correctly, it needs to be extract at the root of the project.

Notebook.ipnyb is the notebook in which we have made our experiments to build our model. All the process is detailled in the notebook.

Training.py and Visualizator.py are the two main scripts of this project.

Training.py is intended to train networks and save the best one in the form of a k-fold cross validation. You can train multiple network structure by adding your own network in the Net module and adding your network class in the net_type array. You can set various parameters like number of epoch, validation split etc... For complete the list of options, please refer to the -h option.

Visualizator.py is intended to see and store results of face recognition on a picture. Options such as threshold, number of minimal votes can be set (Complete list can be found with -h option).

Couple of results can be seen in the results folder. By default, result after visualization are stored in this folder.