Tran Phi Vu

Abstract

Recent years have seen flourishing research on both semi-supervised learningand 3D room layout reconstruction. In this work, we explore the intersection ofthese two fields to advance the research objective of enabling more accurate 3Dindoor scene modeling with less labeled data. We propose the first approach tolearn representations of room corners and boundaries by using a combination oflabeled and unlabeled data for improved layout estimation in a 360-degreepanoramic scene. Through extensive comparative experiments, we demonstrate thatour approach can advance layout estimation of complex indoor scenes using asfew as 20 labeled examples. When coupled with a layout predictor pre-trained onsynthetic data, our semi-supervised method matches the fully supervisedcounterpart using only 12% of the labels. Our work takes an important firststep towards robust semi-supervised layout estimation that can enable manyapplications in 3D perception with limited labeled data.