PBR Textures Capture by CNN Trained in Virtual 3D Scene

Chih Yang

doi:https://doi.org/10.36463/idw.2022.1173

10:20 〜 10:40

[MVS1-1] PBR Textures Capture by CNN Trained in Virtual 3D Scene

*Chih Yang¹、Tzung-Han Lin¹ (1.National Taiwan University of Science and Technology (Taiwan))

physically based rendering, BRDF, deep learning, CNN

https://doi.org/10.36463/idw.2022.1173

This study proposes a PBR (physically based rendering) texture capturing system using CNN (convolutional neural network). To obtain training data, we built a virtual capturing system in computer graphics environment. The final trained CNN was able to generate a set of PBR texture after inputting 4 photos of different lighting conditions.

Proceedings of the International Display Workshops Volume 29 (IDW '22)

[MVS1-1] PBR Textures Capture by CNN Trained in Virtual 3D Scene