CNNFaceRecognition

Built a CNN Model which recognizes people in videos.

Folders structure

All the videos used contained 7 subjects. For each video we extracted the face of each subject in a folder named with the selected subject's unique id. This was done for labeling purposes.

CNN Training, validation and testing

This is contained in Train-Validation-Testing.ipynb We managed single 3-d image data in order to fit CNN and then we used them to train and validate the model. The accuracy was 99% on training, 95% on validation and 86% on test. Test set was completely isolated, during the training session test set example have not been taken under consideration. For each item in test set we printed out a csv file containing actual and predicted value.

In the repository you can find the main plots.

SAVING TRAINED MODEL

This is contained in Train-Validation-Testing.ipynb We saved trained model in files

USING SAVED MODEL TO PREDICT

This is contained in PredictionModule.ipynb We load the best model and use it to make predictions on single files.