Primerjava sistemov fizikalno osnovanih materialov

HIRCI, NEJC (avtor), Bohak, Ciril (mentor)

V okviru diplomske naloge smo analizirali obstoječe sisteme fizikalno osnovanih materialov in nove raziskave na področju ter implementirali svoj dodatek za Blender s tremi različnimi pristopi. Pristop Algorithmic je temeljil na razpoložljivih orodjih v programu Blender, pristopa MaterialGAN in NeuralMaterial pa sta izkoriščala različne modele strojnega učenja. Po implementaciji dodatka smo izpeljali uporabniško študijo, s katero smo želeli oceniti primernost in enostavnost uporabe posameznih pristopov. Tekom študije so sodelujoči izpolnjevali anketo, ki je zajemala standardizirane vprašalnike SUS (angl. System Usability Scale) za ocenjevanje uporabnosti pristopov. Z vsakim pristopom so uporabniki reševali nalogo, v kateri so morali zapolniti pripravljeno 3D sceno z materiali, čim bolj podobnimi materialom v ciljni sceni. Merili smo hitrost in natančnost rešenih nalog ter spremljali komentarje uporabnikov tekom študije. Po analizi rezultatov smo ugotovili, da so se vsi trije pristopi izkazali za izjemno uporabne, a iz različnih razlogov. Pristopa MaterialGAN in NeuralMaterial ponujata dobre razširitve obstoječih pristopov z iterativnim načinom dela in pridobivanjem prepričljivih materialov iz fotografij z bliskavico, pristop Algorithmic pa je mnogo hitrejši in bolj enostaven za uporabo, čeprav v primerjavi z obstoječimi pristopi ponuja manj novih funkcionalnosti.

• Jezik: Slovenski jezik
• Ključne besede: fizikalno osnovani materiali, uporabniška študija, račualniška grafika, fizikalno osnovano upodabljanje, strojno učenje, Blender
• Vrsta gradiva: Diplomsko delo/naloga
• Tipologija: 2.11 - Diplomsko delo
• Organizacija: FRI - Fakulteta za računalništvo in informatiko
• Leto izida: 2022
• PID: 20.500.12556/RUL-138600
• COBISS.SI-ID: 117249283
• Datum objave v RUL: 02.08.2022

HIRCI, NEJC, 2022, Primerjava sistemov fizikalno osnovanih materialov [na spletu]. Diplomsko delo. [Dostopano 14 april 2025]. Pridobljeno s: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=slv&id=138600

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Comparison of physically based material systems

Izvleček

In this thesis, we have analyzed available systems of physically-based materials along with the latest studies in its field. We used the findings of our analysis to develop a Blender add-on with three different approaches for authoring materials. The first approach, Algorithmic, used only available tools in Blender, while the other two approaches, MaterialGAN and NeuralMaterial, used machine learning models. We performed a user study of our implementations, in which we focused on the usability and ease of use of each approach. For this, we used a form, which included the SUS (System Usability Scale) questionnaire. With each of the approaches, users had to perform a task in which they tried to fill a blank 3D scene with materials as similar as possible to those in the target 3D scene. During the study we measured the average task solving time along with the accuracy of user solutions and users' comments about their experience. The results showed that all three approaches have proved extremely useful to the users, but each for different reasons. Both MaterialGAN and NeuralMaterial offered new additions with an iterative workflow and easier extraction of materials directly from flash photos, while Algorithmic was much faster and easier to use, but lacked new worthwile improvements.

• Ključne besede: physically-based materials, user study, computer graphics, physically-based rendering, machine learning, Blender