International Display Workshops General Incorporated Association

10:40 〜 11:00

[MVS1-2] CNN-based Normal Map Generator for Creating Relightable Portrait Images

*Ching-Yi Kuo1、Chih Yang1、Tzung-Han Lin1 (1.National Taiwan University of Science and Technology (Taiwan))

Portrait image relighting, U-net, photometric stereo, normal map

https://doi.org/10.36463/idw.2022.1177

The paper proposes an AI model to generate accurate normal maps for portrait images. We utilized a portrait photo booth system based on photometric methods to generate training data. With these data, users only need to input a portrait image, then it can be converted into normal map image which is further used to relight the color portrait image.