NeRF as super-resolution method

Keywords
Loading...
Thumbnail Image
Issue Date
2021-06-18
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
This thesis is concerning itself with the question whether Neural radiance elds (NeRF) can be used to perform super-resolution. The NeRF method learns a scene representation by receiving images of said scene to train on. As this representation receives information from multiple images we hy- pothesise that the learned representation contains more information than a single image does and could thus create accurate high-resolution outputs while only training on low-resolution images. We observe that the creation of HR images is quite possible, the quality of these is however, only in very limited situations, comparable with interpo- lation based super-resolution methods and signi cantly worse then state-of- the-art methods.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen