Text-to-Image Synthesis with GANs
Explore how to generate realistic images from descriptive text by combining computer vision and natural language processing using GANs. Understand the role of word embeddings, language models, and zero-shot learning in mapping text to images. Gain insight into advanced architectures that enable effective text-to-image synthesis.
We'll cover the following...
In this chapter, we will gradually move from CV to NLP by combining the two fields together, and we will try to generate realistic images from the description text. This process is called text-to-image synthesis (or text-to-image translation).
We know that almost every GAN model generates synthesized data by establishing a definite mapping from a certain form of input data to the output data. Therefore, in order to generate an image from a corresponding description sentence, we need to understand how to represent sentences with vectors.
Quick introduction to word embedding
It is rather easy to define an approach for transforming the words in a sentence into vectors. We can simply assign different ...