Developing Domain-Specific Generative Methods

Abstract

Generative AI is a field that is rapidly developing and growing in scale. As research in this area shifts to building on large-scale foundation models and powerful architectures, careful thought has to go into adapting these models to new domains and tasks. The work in this thesis demonstrates novel approaches to adapting large-scale generative models and architectures to specific applications in virtual try-on, conceptual art, and domain-specific image classification. In addition to the technical contributions, this thesis explores broader open questions about domain-specific generative models; for example, how can we carefully construct our training data to mitigate bias? What do human-in-the-loop methods for creative generative AI look like in practice? To what extent are large-scale vision-language models useful for traditionally image-only tasks?