Semantic Room Layout Estimation from Multiple Views
Title: Semantic Room Layout Estimation from Multiple Views
SNIC Project: Berzelius-2022-205
Project Type: LiU Berzelius
Principal Investigator: David Gillsjö <david.gillsjo@math.lth.se>
Affiliation: Lunds universitet
Duration: 2022-11-01 – 2023-05-01
Classification: 10207
Homepage: https://github.com/DavidGillsjo/SRW-Net
Keywords:

Abstract

Reconstruction of indoor surfaces with limited tex- ture information or with repeated textures, a situation common in walls and ceilings, may be difficult with a monocular Structure from Motion system. Today there exists Neural Networks for Room Layout Estimation which adds valuable information but is exclusively studied for panorama images. The goal of the project is to develop a new representation of Room Layout and corresponding Neural Network for detection. This new representation will contain more semantic information, flexibility and allows for generalization to a Multi-View problem statement. A Multi-View formulation is more useful since it can be applied to any Structure from Motion pipeline.