Estimating Illumination for Augmented Reality Dental Application Using Deep Learning
Title: Estimating Illumination for Augmented Reality Dental Application Using Deep Learning
SNIC Project: LiU-gpu-2021-2
Project Type: LiU Compute
Principal Investigator: George Osipov <george.osipov@liu.se>
Affiliation: Linköpings universitet
Duration: 2021-02-03 – 2021-09-01
Classification: 10207
Keywords:

Abstract

This project is part of the Master thesis at the Linköping university. In this master thesis project, a model for estimating illumination, using Deep learning, will be developed and optimised for the purpose of realistically rendering superimposed teeth for use with mobile augmented reality. Today, tools such as AR-kit can estimate lighting, however, this lighting does not properly take into account the in-door our out-door lighting and as such provides very poor lighting estimation, consequently unrealistic superimposition of 3D objects. Our project will attempt to estimate accurate illumination from a single LDR image, taken with a phone, with a large part of the screen real-estate being occupied by a face, in real-time. Many Deep learning models today are not optimised for mobile phones, consequently performing poorly in terms of inference time and requiring a lot CPU processing power.