Text to Image Generation Using Gan

Text to image synthesis, one of the most fascinating applications of GANs, is one of the hottest topics in all of machine learning and artificial intelligence. This paper comprises techniques for training a GAN to synthesise human faces and images of flowers from text descriptions. In this paper, we are proposing to train the GAN progressively as proposed in the ProGAN architecture and along with that trying to improve its results by proposing a custom update rule for alpha which controls the fading rate during the progressive growth of the architecture. With experimental testing using the Oxford102 and LFW datasets, our proposed architecture and training process ensures fast learning and smooth transitions between each trained generation.