Automatic 3D reconstruction of a room's bounding surfaces from a single image is a very active research topic with many practical applications, which include the generation or revision of building information models (BIM), the support in the creation of digital twins of buildings for optimization and monitoring purposes, and many others.
The Visual Computing Group of CRS4 is very active in this area, and has recently developed a novel end-to-end approach to predict a 3D room layout from a single panoramic image. The new method, based on a new neural network architecture, outperforms state-of-the-art solutions in prediction accuracy, in particular in cases of complex wall layouts or curved wall footprints.
The work will be presented by Giovanni Pintore at the forthcoming European Conference on Computer Vision (ECCV), which will be held from 23 to 28 August 2020. ECCV is the major annual event in the areas of computer vision and pattern recognition. This year's edition, due to COVID-19 restrictions, will be held online. All the material will remain accessible to participants on demand until June 2021.
All the details of the method are described in the following article:
Giovanni Pintore, Marco Agus, and Enrico Gobbetti. AtlantaNet: Inferring the 3D Indoor Layout from a Single 360 Image beyond the Manhattan World Assumption. In Proc. ECCV, August 2020. To appear.
Article Preprint.
The research has received funding from Sardinian Regional Authorities under projects VIGECLAB, AMAC, and TDM (POR FESR 2014-2020). We also acknowledge the contribution of the European Union's H2020 research and innovation programme under grant agreements 813170 (EVOCATION).