Controlled Face Generation Using Conditional GAN

The Conditional GAN generates the human faces with certain conditions applied to its features. In Conditional GANs, the generator and discriminator networks are extended to take in additional conditional information during the training process. The generator network synthesizes faces with specific attributes. The discriminator is also conditioned on the same external information which helps it learn not only to differentiate between real and fake samples but also to recognize and potentially control specific attributes in the generated data. By altering the conditionally labelled data supplied to the discriminator and generator, the characteristics of the created faces can be modified. The conditional attributes include hair color, gender, face shape, expression, and age. Applying the conditions while keeping the random noise constant leads to generating faces that exhibit those specific attributes. The convergence occurs while the discriminator can no longer reliably distinguish between real and fake images. This point indicates that the generator has successfully learned to produce realistic-looking images that resemble the real data distribution. The trained generator network can be used to create new, artificial faces when the GAN has converged.