There’s something about your face. Exploring Generative Adversarial Networks as the basis for examining the memorability of human faces
Keywords
Loading...
Authors
Issue Date
2021-01-24
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
What makes a face memorable? That is a complicated and challenging question. Several studies so far came up with indeterminate and non-specific results. GANs[1], which are potent tools using deep learning to generate photo-realistic images artificially, will be used to address this question. Using the face dataset[4] annotated by humans with memorability scores faces images grouped into two classes; high-memorable and low-memorable. Using GANs and how they encode the images - as the latent vectors, it will be possible to obtain a vector directed towards high-memorable latent vector space. It is hypothesized that using this vector it will be possible to make the face images more memorable. Here, in this work, the process to obtain such a vector and validate experimentally that it indeed is capable of making faces more memorable will be explained thoroughly. The results might pave the way towards algorithms or approaches that manipulate the facial memorability. On the other hand, it could help understand and further develop the research concerning this phenomenon.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen