8 faces FCN
In this exercise we work with the 8 faces dataset. this dataset has 350 images of 8 celebrities.
To get an overview of the data open the notebook 8 faces overview and look at the images.
The data is a random sample of 8 persons of the OXFORD VGG Face dataset (over 2600 Persons),
for more information look here: http://www.robots.ox.ac.uk/~vgg/data/vgg_face/
a) Have a look at the 8_faces_fc notebook and try to understand it.
In this notebook we traind a fully connected neural network on the dataset. The accuracy of this network on the validation data is ~60%.
You can also see how to save and reload a trained model in keras, you will have to do that for the next task.
8 faces your own model CNN
b) Now it’s your turn!
Design a network that outperforms this baseline fc model.
You can use the test data the check how good your model performs on new unseen data.
Hint 1: Is it a good idea to use a fully connected neural network on this dataset?
Hint 2 : The training of more complex networks could take some time because we compute only on cpu. (up to 1h)
Hint 3 : Look at the possible solution to get some ideas 8_faces_cnn