Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scaled 360 layouts: Revisiting non-central panoramas

Bruno Berenguel-Baeta, Jesus Bermudez-Cameo, Jose J. Guerrero

TL;DR

The paper addresses the challenge of recovering scaled 3D indoor layouts from a single non-central panorama, where scale cues are typically missing. It introduces a pipeline that first extracts structural boundary lines with a HorizonNet‑based network adapted for non-central circular panoramas, then uses two DL T‑like linear solvers to recover room height and wall positions under Manhattan and Atlanta world assumptions. A synthetic non-central panorama dataset supports training, and the approach achieves superior 3D layout accuracy while recovering scale without priors, outperforming state-of-the-art methods. This enables robust, scale‑correct indoor scene understanding from a single panoramic image, with practical implications for 3D reconstruction and navigation tasks.

Abstract

From a non-central panorama, 3D lines can be recovered by geometric reasoning. However, their sensitivity to noise and the complex geometric modeling required has led these panoramas being very little investigated. In this work we present a novel approach for 3D layout recovery of indoor environments using single non-central panoramas. We obtain the boundaries of the structural lines of the room from a non-central panorama using deep learning and exploit the properties of non-central projection systems in a new geometrical processing to recover the scaled layout. We solve the problem for Manhattan environments, handling occlusions, and also for Atlanta environments in an unified method. The experiments performed improve the state-of-the-art methods for 3D layout recovery from a single panorama. Our approach is the first work using deep learning with non-central panoramas and recovering the scale of single panorama layouts.

Scaled 360 layouts: Revisiting non-central panoramas

TL;DR

The paper addresses the challenge of recovering scaled 3D indoor layouts from a single non-central panorama, where scale cues are typically missing. It introduces a pipeline that first extracts structural boundary lines with a HorizonNet‑based network adapted for non-central circular panoramas, then uses two DL T‑like linear solvers to recover room height and wall positions under Manhattan and Atlanta world assumptions. A synthetic non-central panorama dataset supports training, and the approach achieves superior 3D layout accuracy while recovering scale without priors, outperforming state-of-the-art methods. This enables robust, scale‑correct indoor scene understanding from a single panoramic image, with practical implications for 3D reconstruction and navigation tasks.

Abstract

From a non-central panorama, 3D lines can be recovered by geometric reasoning. However, their sensitivity to noise and the complex geometric modeling required has led these panoramas being very little investigated. In this work we present a novel approach for 3D layout recovery of indoor environments using single non-central panoramas. We obtain the boundaries of the structural lines of the room from a non-central panorama using deep learning and exploit the properties of non-central projection systems in a new geometrical processing to recover the scaled layout. We solve the problem for Manhattan environments, handling occlusions, and also for Atlanta environments in an unified method. The experiments performed improve the state-of-the-art methods for 3D layout recovery from a single panorama. Our approach is the first work using deep learning with non-central panoramas and recovering the scale of single panorama layouts.
Paper Structure (6 sections, 8 equations, 4 figures, 1 table)

This paper contains 6 sections, 8 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Proposed method
  3. Neural network as line extractor
  4. Geometric solvers
  5. Experiments
  6. Conclusions

Figures (4)

  • Figure 1: Central (up-left) and non-central panoramas (bottom-left) have similar appearance but there are subtle differences in favor of the second if we want to obtain 3D information including the scale. On the right, the scaled layout obtained from a non-central panorama in an Atlanta environment with our solution.
  • Figure 2: Pipeline of the proposed method. The non-central circular panorama is processed by the fine-tuned network. The network provides the pixel information of the structural lines and a per-column probability of a wall-wall intersection. Then the proposed geometric pipeline, including the new solvers, gives the final scaled layout.
  • Figure 3: Rays and wall parameter definition. The parameters are: wall reference system $\{\mathbf{e}_1,\mathbf{e}_2,\mathbf{e}_3\}$; $\boldsymbol{\Xi}$ and $\boldsymbol{X}$ define the projecting rays; $\mathbf{(l,\bar{l})}$ and $\mathbf{(m,\bar{m})}$ are the ceiling and floor lines that define the wall; $\mathbf{x_L,x_M}$ define the closest points of the lines to the origin; $h_c$, $h_f$ and $d$ are the ceiling and floor height and distance to the wall respectively.
  • Figure 4: Examples of 3D reconstruction from the proposed method. We shown the non-central panorama and the 3D reconstruction. The green wire frame is the real 3D layout of the room.