Image synthesis for machine learning

Dnr:

SNIC 2017/5-20

Type:

SNAC Small

Principal Investigator:

Jonas Unger

Affiliation:

Linköpings universitet

Start Date:

2017-03-20

End Date:

2018-04-01

Primary Classification:

10207: Datorseende och robotik (autonoma system)

Webpage:

Allocation

Abstract

In this project we are investigating how image synthesis and computer graphics can be used to generate labeled and segmented training data for deep neural networks. The goal is to investigate how real manually labeled training data can be replaced by synthetic data. The motivation behind the project is that hand labeling of training data is very time consuming and prone to inaccuracies and errors.