Optimized Mirror Generative Adversarial Network with BERT Neural Architecture for Text Caption to Image Conversion

In the past few years, there have been many advancements in the field of Generative Adversarial Networks (GANs). The paper talks about the various types of GANs developed along with focusing on one specific application of generating a human face using the given text description which is a less-explored area. GANs are a class of machine learning models designed for generative tasks, such as creating realistic images, music, or text. GANs are a powerful tool in the field of deep learning. This paper explains StackGAN, AttentionalGAN, MirrorGAN, CycleGAN, etc. Beyond this, the paper also comprises the various embedding techniques, their advantage, models, and disadvantages. Through this paper, we also got insight into how to improve the performance of models just by improving the embeddings or by pre-training the models in the case of MirrorGAN.