Abstract
We present a pipeline for generating defurnished replicas of indoor spacesrepresented as textured meshes and corresponding multi-view panoramic images.To achieve this, we first segment and remove furniture from the meshrepresentation, extend planes, and fill holes, obtaining a simplifieddefurnished mesh (SDM). This SDM acts as an ``X-ray'' of the scene's underlyingstructure, guiding the defurnishing process. We extract Canny edges from depthand normal images rendered from the SDM. We then use these as a guide to removethe furniture from panorama images via ControlNet inpainting. This controlsignal ensures the availability of global geometric information that may behidden from a particular panoramic view by the furniture being removed. Theinpainted panoramas are used to texture the mesh. We show that our approachproduces higher quality assets than methods that rely on neural radiancefields, which tend to produce blurry low-resolution images, or RGB-Dinpainting, which is highly susceptible to hallucinations.