/realistic-aerial-imagery-style-transfer-bsc

Repository of my bachelors diploma thesis titled "Translating the style of aerial imagery of generated landscapes into a realistic look"

Primary LanguagePythonMIT LicenseMIT

Style transfer of computer-generated orthophoto aerial images into a realistic look

A repository containing code developed as part of the Bachelor thesis at the University of Ljubljana, Faculty of Computer and Information Science by Nejc Krajšek.

This repository contains the code, data, and results of the project. The project uses style transfer to translate the style of aerial imagery of generated landscapes into a realistic look. The code is implemented in Python using the PyTorch library. The data consists of aerial imagery of real landscapes and generated landscapes. The results show that the style transfer technique is able to produce realistic aerial imagery of generated landscapes.

Abstract

Neural Style Transfer is an interesting and promising approach in the field of computer vision and image processing. The technique, based on deep neural networks, allows the content of one image to be fused with the style of another image to produce visually striking and compelling images. While the use of Neural Style Transfer has been extensively explored in various fields including artistic image creation and photo manipulation, its potential in the field of orthophoto imagery has been a relatively unexplored area.

Orthophoto imagery, with its top-down view and geometric accuracy, is an invaluable asset in fields such as cartography, urban planning, agriculture and environmental monitoring, as well as in creating terrain in virtual worlds. However, the production of orthophotos in a standard way requires a lot of time and equipment, so the use of neural style transfer in this field has enormous potential.

The thesis is accessible at: [https://repozitorij.uni-lj.si/Dokument.php?id=174487]

Bibtex:

@thesis{Krajsek2023,
    title = {Style transfer of computer-generated orthophoto aerial images into a realistic look},
    author = {Krajšek, Nejc},
    year = {2023},
    school = {University of Ljubljana, Faculty of Computer and Information Science},
    type = {Bachelor thesis},
    address = {Ljubljana, Slovenia},
    note = {Mentor: Ciril Bohak, Language: Slovenian, Slovenian title: Prenos stila računalniško generiranih ortofoto posnetkov pokrajine v realistični izgled},
    url = {https://repozitorij.uni-lj.si/Dokument.php?id=174487}
}