InLayDiffusion: Indoor Layout Estimation from a Single Panorama via Structural Point Cloud Lifting and Guided Diffusion
Giovanni Pintore, Uzair Shah, Marco Agus, and Enrico Gobbetti
2026
Abstract
We introduce a novel end-to-end deep-learning method that combines structural geometric lifting with prior-guided diffusion to recover a 2D floorplan and a 3D room layout from a 360-degree image. Structural lifting predicts gravity-aligned, structure-aware depth and projects the resulting colored 3D point cloud onto the floor plane through a differentiable operation, directly linking panoramic image features to geometric reasoning and improving robustness to clutter while encoding cues such as room height and coarse footprint geometry. Footprint reconstruction is then formulated as polygon denoising and completion within a diffusion framework. The layout is represented as a fixed-vertex polygon jointly encoded with floor-projected RGB and density features, enabling refinement of noisy predictions and completion of occluded boundaries using structural priors. A variational guidance network further regularizes initialization by encoding indoor layout priors. Panoramic benchmarks show that our diffusion approach provides a strong foundation for general room reconstruction, even in cluttered, non-Manhattan environments.
Reference and download information
Giovanni Pintore, Uzair Shah, Marco Agus, and Enrico Gobbetti. InLayDiffusion: Indoor Layout Estimation from a Single Panorama via Structural Point Cloud Lifting and Guided Diffusion. In Computer Graphics International. Volume 10605 of Lecture Notes in Computer Science (LNCS), Springer, 2026.
Related multimedia productions
Bibtex citation record
@incollection{Pintore:2026:IIL, author = {Giovanni Pintore and Uzair Shah and Marco Agus and Enrico Gobbetti}, title = {{InLayDiffusion}: Indoor Layout Estimation from a Single Panorama via Structural Point Cloud Lifting and Guided Diffusion}, booktitle = {Computer Graphics International}, series = {Lecture Notes in Computer Science (LNCS)}, volume = {10605}, publisher = {Springer}, year = {2026}, abstract = { We introduce a novel end-to-end deep-learning method that combines structural geometric lifting with prior-guided diffusion to recover a 2D floorplan and a 3D room layout from a 360-degree image. Structural lifting predicts gravity-aligned, structure-aware depth and projects the resulting colored 3D point cloud onto the floor plane through a differentiable operation, directly linking panoramic image features to geometric reasoning and improving robustness to clutter while encoding cues such as room height and coarse footprint geometry. Footprint reconstruction is then formulated as polygon denoising and completion within a diffusion framework. The layout is represented as a fixed-vertex polygon jointly encoded with floor-projected RGB and density features, enabling refinement of noisy predictions and completion of occluded boundaries using structural priors. A variational guidance network further regularizes initialization by encoding indoor layout priors. Panoramic benchmarks show that our diffusion approach provides a strong foundation for general room reconstruction, even in cluttered, non-Manhattan environments. }, url = {http://vic.crs4.it/vic/cgi-bin/bib-page.cgi?id='Pintore:2026:IIL'}, }
The publications listed here are included as a means to ensure timely
dissemination of scholarly and technical work on a non-commercial basis.
Copyright and all rights therein are maintained by the authors or by
other copyright holders, notwithstanding that they have offered their works
here electronically. It is understood that all persons copying this
information will adhere to the terms and constraints invoked by each
author's copyright. These works may not be reposted without the
explicit permission of the copyright holder.
Please contact the authors if you are willing to republish this work in
a book, journal, on the Web or elsewhere. Thank you in advance.
All references in the main publication page are linked to a descriptive page
providing relevant bibliographic data and, possibly, a link to
the related document. Please refer to our main
publication repository page for a
page with direct links to documents.
+39 070 9250 1
+39 070 9250 216
(Secretary) katia
crs4.it
www.crs4.it/vic/
