What is STACKGAN

A picture is worth a thousand words goes the well-known adage. Generating images from text understandably has many uses. In this chapter, the authors explore a state-of-the-art generative deep learning method to produce synthetic images and a new better way for evaluating the same. The approach focuses on synthesizing high-resolution images with multiple objects present in an image, given the textual description of the images. The existing literature uses object pathway GAN (OP-GAN) to automatically generate images from text. The work described in this chapter attempts to improvise the discriminator network from the original implementation using OP-GAN. This eventually helps the generator network's learning rate adjustment based on the discriminator output. Finally, the trained model is evaluated using semantic object accuracy (SOA), the same metric that is used to evaluate the baseline implementation, which is better than the metrics used previously in the literature.