Conditional generative adversarial nets for convolutional face generation
J.Gauthier, Stanford 2014.
This project aims in developing C-GANs(Conditional Generative adversarial nets) for face generation. Making it clear what the generator and discriminator does exactly in GAN. The G networks task is to fool D network by making it to believe that the samples are real data and is not generated by G.
D is trained to maximize the probability of training data and to minimize the probability of data sampled from G. Both D and G are MLPs and trained using SGD.
C-GANs adds conditioning (y) to both G and D. This y can be any attributes (face) or labels.In this way the faces can be generated and it is estimated by using parzen window.
D is trained to maximize the probability of training data and to minimize the probability of data sampled from G. Both D and G are MLPs and trained using SGD.
C-GANs adds conditioning (y) to both G and D. This y can be any attributes (face) or labels.In this way the faces can be generated and it is estimated by using parzen window.