Primerjava metod za ocenjevanje polja sevanja

Peršak, Vid (author), Čehovin Zajc, Luka (mentor)

Problem digitalnega opisa 3D sveta obstaja že od začetkov področja računalniške grafike. Večina pristopov temelji na rekonstrukciji sveta iz množice fotografij iste scene. Najnovejše metode temeljijo na globokem učenju, ki omogoča neposredno ocenjevanje polja sevanja. Nadaljnji razvoj metod izboljšuje hitrost, natančnost in dostopnost. Cilj diplomske naloge je pregled področja ter primerjava izbranih metod za ocenjevanje polja sevanja. V eksperimentalni analizi ovrednotimo kvaliteto metod, njihovo odvisnost od ločljivosti in števila vhodnih slik ter njihove potrebe po računskih virih.

• Language: Slovenian
• Keywords: Globoko učenje, Nevronske mreže, 3D rekonstrukcija, NeRF, Gaussovo Packanje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-160439
• COBISS.SI-ID: 208519939
• Publication date in RUL: 28.08.2024

PERŠAK, Vid, 2024, Primerjava metod za ocenjevanje polja sevanja [online]. Bachelor’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=160439

Secondary language

• Language: English
• Title: Comparison of radiance field estimation methods

Abstract

The problem of digitally describing a 3D world has existed since the beginnings of computer graphics. Most approaches are based on the reconstruction of a world from a set of photographs of the same scene. The latest methods are based on deep learning, which allows for direct estimation of radiance fields. The subsequent development of these methods increases speed, accuracy and accessibility. The goal of this thesis is to review the field and to compare the chosen methods for radiance field estimation. In the experimental analysis, we evaluate the quality of the methods, their dependence on resolution, the number of input images and their computational resource requirements.

• Keywords: Deep learning, Neural networks, 3D reconstruction, NeRF, Gaussian Splatting