Zhao Yining ; Wen Chao ; Xue Zhou ; Gao Yue

Abstract

Significant geometric structures can be compactly described by globalwireframes in the estimation of 3D room layout from a single panoramic image.Based on this observation, we present an alternative approach to estimate thewalls in 3D space by modeling long-range geometric patterns in a learnableHough Transform block. We transform the image feature from a cubemap tile tothe Hough space of a Manhattan world and directly map the feature to thegeometric output. The convolutional layers not only learn the localgradient-like line features, but also utilize the global information tosuccessfully predict occluded walls with a simple network structure. Unlikemost previous work, the predictions are performed individually on each cubemaptile, and then assembled to get the layout estimation. Experimental resultsshow that we achieve comparable results with recent state-of-the-art inprediction accuracy and performance. Code is available athttps://github.com/Starrah/DMH-Net.