Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unit Region Encoding: A Unified and Compact Geometry-aware Representation for Floorplan Applications

Huichao Zhang, Pengyu Wang, Manyi Li, Zuojun Li, Yaguang Wu

TL;DR

This work introduces Unit Region Encoding (URE), a boundary-adaptive, geometry-aware representation for floorplans that partitions irregular layouts into unit regions guided by a density map. A two-module network, URE-Net, learns region-wise latent encodings from the density map and floorplan masks, enabling flexible application across interior space planning, floorplan metric learning, and floorplan generation. The density map is defined as $f(x)=1/(d_N(x)+d_S(x)+d_E(x)+d_W(x))$ for $x\in\Omega$ and $0$ outside, guiding the partition into meaningful regions. Extensive experiments and ablations demonstrate that unit-region encodings improve accuracy and visual quality over raster and graph-based representations, offering a compact, geometry-aware alternative for diverse floorplan tasks with practical implications for design and analysis.

Abstract

We present the Unit Region Encoding of floorplans, which is a unified and compact geometry-aware encoding representation for various applications, ranging from interior space planning, floorplan metric learning to floorplan generation tasks. The floorplans are represented as the latent encodings on a set of boundary-adaptive unit region partition based on the clustering of the proposed geometry-aware density map. The latent encodings are extracted by a trained network (URE-Net) from the input dense density map and other available semantic maps. Compared to the over-segmented rasterized images and the room-level graph structures, our representation can be flexibly adapted to different applications with the sliced unit regions while achieving higher accuracy performance and better visual quality. We conduct a variety of experiments and compare to the state-of-the-art methods on the aforementioned applications to validate the superiority of our representation, as well as extensive ablation studies to demonstrate the effect of our slicing choices.

Unit Region Encoding: A Unified and Compact Geometry-aware Representation for Floorplan Applications

TL;DR

This work introduces Unit Region Encoding (URE), a boundary-adaptive, geometry-aware representation for floorplans that partitions irregular layouts into unit regions guided by a density map. A two-module network, URE-Net, learns region-wise latent encodings from the density map and floorplan masks, enabling flexible application across interior space planning, floorplan metric learning, and floorplan generation. The density map is defined as for and outside, guiding the partition into meaningful regions. Extensive experiments and ablations demonstrate that unit-region encodings improve accuracy and visual quality over raster and graph-based representations, offering a compact, geometry-aware alternative for diverse floorplan tasks with practical implications for design and analysis.

Abstract

We present the Unit Region Encoding of floorplans, which is a unified and compact geometry-aware encoding representation for various applications, ranging from interior space planning, floorplan metric learning to floorplan generation tasks. The floorplans are represented as the latent encodings on a set of boundary-adaptive unit region partition based on the clustering of the proposed geometry-aware density map. The latent encodings are extracted by a trained network (URE-Net) from the input dense density map and other available semantic maps. Compared to the over-segmented rasterized images and the room-level graph structures, our representation can be flexibly adapted to different applications with the sliced unit regions while achieving higher accuracy performance and better visual quality. We conduct a variety of experiments and compare to the state-of-the-art methods on the aforementioned applications to validate the superiority of our representation, as well as extensive ablation studies to demonstrate the effect of our slicing choices.
Paper Structure (11 sections, 3 equations, 6 figures, 5 tables)

This paper contains 11 sections, 3 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Floorplan Representation
  4. Density Map
  5. Unit Region Partition
  6. URE-Net architecture
  7. Applications
  8. Interior Space Planning
  9. Floorplan Metric Learning
  10. Floorplan Generation.
  11. Conclusion

Figures (6)

  • Figure 1: The visualization of the proposed density map and unit region partition under different splitting strategies. The last picture shows the uniform partition without considering the particular floorplan shape, causing the misalignment between the regions and the floorplan boundary and thus the irregular regions.
  • Figure 2: The statistics and a visualization of the collected functional area annotation on 3D-FRONT dataset fu20213d. We consider the four majority types and merge the rest of them as "others", forming five categories in our setting.
  • Figure 3: Qualitative comparison of the interior space planning application. The results are produced by DeepLabv3+ network (pixel-wise) and URE-Net+DeepLabv3+ (region-wise).
  • Figure 4: The floorplan retrieval comparison between our method and LayoutGMN patil2021layoutgmn.
  • Figure 5: The ground-truth floorplan and the generated results of our method and Graph2Plan hu2020graph2plan.
  • ...and 1 more figures