Modelování obousměrných distribučních funkcí odrazivosti pomocí neuronových sítí

Abstract

In this thesis, I explore the modeling of bidirectional reflectance distribution functions using neural networks. The aim is to design and implement neural networks that can efficiently approximate the BRDF of materials and represent a certain BRDF model. To achieve this goal, I have designed and tested various neural network architectures for representing test BRDF models. Issues one might encounter in the design and training of these networks have been discussed. I then compared the results with their analytical counterparts, compared rendering speeds, and validated their accuracy using the furnace test. The achieved results show that the proposed neural networks achieve a very high level of accuracy in rendering, confirming their potential for describing the reflective properties of materials for the purposes of synthesizing photorealistic images.