Deep neural network (DNN) has been successfully used to generate face images of only database subjects. Can we use DNN to generate face image of a novel person outside the database?
Hello Wasseem,
Have you ever try to generate new faces based on your database using GAN (Generative Adversarial Network)?
Maybe this paper is closed to this case.
https://pdfs.semanticscholar.org/42f6/f5454dda99d8989f9814989efd50fe807ee8.pdf
Best regards.
Yes. GANs seem to do this all the time. However, whether the GANs can only generate faces which are formed by mixing the characteristics of known faces in the training set or can generate faces with new unique features (while keeping the faces human) is hard to tell. I believe that GANs are more inclined towards the former property, i.e. generating faces consisting of novel mixes of the known facial features from the training set.
Hi, I would also recommend to take a look at variational autoencoders, as an alternative to GAN.
https://towardsdatascience.com/intuitively-understanding-variational-autoencoders-1bfe67eb5daf
- Badges
- Science topic