Corners for Layout: End-to-End Layout Recovery from 360 Images

The problem of 3D layout recovery in indoor scenes has been a core researchtopic for over a decade. However, there are still several major challenges thatremain unsolved. Among the most relevant ones, a major part of thestate-of-the-art methods make implicit or explicit assumptions on the scenes --e.g. box-shaped or Manhattan layouts. Also, current methods are computationallyexpensive and not suitable for real-time applications like robot navigation andAR/VR. In this work we present CFL (Corners for Layout), the first end-to-endmodel for 3D layout recovery on 360 images. Our experimental results show thatwe outperform the state of the art relaxing assumptions about the scene and ata lower cost. We also show that our model generalizes better to camera positionvariations than conventional approaches by using EquiConvs, a type ofconvolution applied directly on the sphere projection and hence invariant tothe equirectangular distortions. CFL Webpage: https://cfernandezlab.github.io/CFL/