Exploring Transformers and Degradation Methods in the Super Resolution Field

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2022-12-12

Department

Major/Subject

Machine Learning, Data Science and Artificial Intelligence

Mcode

SCI3044

Degree programme

Master's Programme in Computer, Communication and Information Sciences

Language

en

Pages

63+32

Series

Abstract

Super Resolution is one of the most difficult fields to explore as the real world degradations are unknown and hard to be mathematically modeled. This research project aims at exploring different approaches for improving both efficiency and results of the existing algorithms by adapting a denoising method for the Super Resolution task and implementing a new degradation pipeline which would better simulate the real scenarios. The method was evaluated on three datasets containing reference images and performs the best on average. For real images which do not contain a reference, our solution provides results with more details and textures, therefore having a more pleasant looking outcome.

Description

Supervisor

Jung, Alexander

Thesis advisor

Yu, Tian
Yamac, Mehmet

Keywords

Super resolution, transformers, degradations, encoder-decoder

Other note

Citation