Real image transformation through conditional GANs

Generative Adversarial Networks (GANs) has become a new big topic as they are able to produce diverse high-resolution and photo-realistic images, often indistinguishable from real images and this ability has powered many real-world applications. However, these networks are limited on generating from...

Descripción completa

Detalles Bibliográficos
Autor: Grau Gasulla, Martí
Tipo de recurso: tesis de maestría
Fecha de publicación:2021
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/343954
Acceso en línea:https://hdl.handle.net/2117/343954
Access Level:acceso embargado
Palabra clave:deep learning
Computer Vision
GANs
decoration
Descripción
Sumario:Generative Adversarial Networks (GANs) has become a new big topic as they are able to produce diverse high-resolution and photo-realistic images, often indistinguishable from real images and this ability has powered many real-world applications. However, these networks are limited on generating from random samples (latent vectors) with the image domain trained for. The interest of this thesis is to be able to adapt this type of models and increase the functionality of them by transforming real images through attribute conditions.