Abstract

Generative machine learning models have achieved unprecedented feats in recent years and look primed to reach even more impressive heights. By learning data distributions through unsupervised training and by leveraging the power of neural networks, these models are responsible for breakthroughs in various domains. The aim of this paper is to cover some of the prominent generative model architectures through the bottom-up construction of an illustrated storybook generating interface that uses transfer learning on a transformer-based text generator, and the Vector Quantized Generative Adversarial network (VQGAN) coupled with Contrastive Language–Image Pre-training (CLIP) for prompt-driven image generation.

Advisor

Chowdhury, Subhadip

Second Advisor

Bhowmik, Khowshik

Department

Computer Science; Mathematics

Disciplines

Art and Design | Computer Engineering

Keywords

Transformers, GANs, VQGAN, text-to-image synthesis, image generation, text generation

Publication Date

2023

Degree Granted

Bachelor of Arts

Document Type

Senior Independent Study Thesis

Share

COinS
 

© Copyright 2023 Ussama Mustafa