StyleGAN and DCGAN for Face Generation: A Comparative Evaluation

Advent of GAN networks has enabled several tasks such as text to face generation easier. It helps in synthesizing several instances of data from the actual data. It gives an idea on new possibilities for existing dataset. The current scenario in literature sees a parade of GAN architectures. The performance of each GAN architecture varies in a wide range of outputs. Hence, there always exists a confusion of the GAN architecture to be chosen for the purpose of the task in question. This paper attempts to make a comparative evaluation of two famous GAN architectures, namely, StyleGAN and DCGAN. The architectures are implemented and the images generated are compared on the basis of the quality of images. It is seen that StyleGAN responds more accurately to features like smile, age, and gender. From the experiments, it is seen that the FID score of StyleGAN has a lower value as compared to that of DCGAN.