Conditioning diffusion-based generative models with semantical binary attribute descriptors

In this thesis, we present a novel architecture that enables image generation by dataset-specific binary attributes. The resulting pipeline has been part of a larger research project done between Northeastern University and UPenn. This thesis describe the main contributions to it, and the rationale...

ver descrição completa

Detalhes bibliográficos
Autor: Formosa Marin, Feliu
Tipo de documento: dissertação
Data de publicação:2025
País:España
Recursos:Universitat Politècnica de Catalunya (UPC)
Repositório:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglês
OAI Identifier:oai:upcommons.upc.edu:2117/452213
Acesso em linha:https://hdl.handle.net/2117/452213
Access Level:Acceso aberto
Palavra-chave:Computer vision
Deep learning
Diffusion models
Encodings
Stable Diffusion
Binary attributes
Visió per ordinador
Aprenentatge profund
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic
Descrição
Resumo:In this thesis, we present a novel architecture that enables image generation by dataset-specific binary attributes. The resulting pipeline has been part of a larger research project done between Northeastern University and UPenn. This thesis describe the main contributions to it, and the rationale behind its design pieces.