We are computer vision PhD students so that is our expertise. Probably also fake awareness
What it does
It artificially alters images of female faces (no make-up) such that after the altering process make-up has been applied on the face.
How we built it
We use 2 neural networks competing against each other. One neural net generates fakes and the other detects it. After a long training process the generator becomes so well at faking good looking images that the detector isn't able to tell fake from real.
Challenges we ran into
Mostly convergence issues and finding the right parameters for the learning process.
Accomplishments that we're proud of
Artificially generating fake images where most people would not notice. We farmed 200K images of celebrities that we used to train the network for makeupifycation
What we learned
Data is one of the most important part.
What's next for DeepFakes
Do actually something meaningful -> Beautyfication?