Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoofDiffusion: Constructing Roofs from Severely Corrupted Point Data via Diffusion

Kyle Shih-Huang Lo, Jörg Peters, Eric Spellman

TL;DR

Tested with the leading City3D algorithm, preprocessing height maps with RoofDiffusion noticeably improves 3D building reconstruction and quantitatively outperforms state-of-the-art unguided depth completion and representative inpainting methods for Digital Elevation Models (DEM).

Abstract

Accurate completion and denoising of roof height maps are crucial to reconstructing high-quality 3D buildings. Repairing sparse points can enhance low-cost sensor use and reduce UAV flight overlap. RoofDiffusion is a new end-to-end self-supervised diffusion technique for robustly completing, in particular difficult, roof height maps. RoofDiffusion leverages widely-available curated footprints and can so handle up to 99\% point sparsity and 80\% roof area occlusion (regional incompleteness). A variant, No-FP RoofDiffusion, simultaneously predicts building footprints and heights. Both quantitatively outperform state-of-the-art unguided depth completion and representative inpainting methods for Digital Elevation Models (DEM), on both a roof-specific benchmark and the BuildingNet dataset. Qualitative assessments show the effectiveness of RoofDiffusion for datasets with real-world scans including AHN3, Dales3D, and USGS 3DEP LiDAR. Tested with the leading City3D algorithm, preprocessing height maps with RoofDiffusion noticeably improves 3D building reconstruction. RoofDiffusion is complemented by a new dataset of 13k complex roof geometries, focusing on long-tail issues in remote sensing; a novel simulation of tree occlusion; and a wide variety of large-area roof cut-outs for data augmentation and benchmarking.

RoofDiffusion: Constructing Roofs from Severely Corrupted Point Data via Diffusion

TL;DR

Tested with the leading City3D algorithm, preprocessing height maps with RoofDiffusion noticeably improves 3D building reconstruction and quantitatively outperforms state-of-the-art unguided depth completion and representative inpainting methods for Digital Elevation Models (DEM).

Abstract

Accurate completion and denoising of roof height maps are crucial to reconstructing high-quality 3D buildings. Repairing sparse points can enhance low-cost sensor use and reduce UAV flight overlap. RoofDiffusion is a new end-to-end self-supervised diffusion technique for robustly completing, in particular difficult, roof height maps. RoofDiffusion leverages widely-available curated footprints and can so handle up to 99\% point sparsity and 80\% roof area occlusion (regional incompleteness). A variant, No-FP RoofDiffusion, simultaneously predicts building footprints and heights. Both quantitatively outperform state-of-the-art unguided depth completion and representative inpainting methods for Digital Elevation Models (DEM), on both a roof-specific benchmark and the BuildingNet dataset. Qualitative assessments show the effectiveness of RoofDiffusion for datasets with real-world scans including AHN3, Dales3D, and USGS 3DEP LiDAR. Tested with the leading City3D algorithm, preprocessing height maps with RoofDiffusion noticeably improves 3D building reconstruction. RoofDiffusion is complemented by a new dataset of 13k complex roof geometries, focusing on long-tail issues in remote sensing; a novel simulation of tree occlusion; and a wide variety of large-area roof cut-outs for data augmentation and benchmarking.
Paper Structure (35 sections, 6 equations, 22 figures, 7 tables, 3 algorithms)

This paper contains 35 sections, 6 equations, 22 figures, 7 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Statement
  4. RoofDiffusion
  5. Roof-Focused Height Map Normalization
  6. Height Completion based on a Diffusion Model
  7. Datasets & Benchmarks
  8. Tree Points
  9. Incompleteness
  10. Benchmarks
  11. Experiments
  12. Quantitative Evaluation of Height Completion
  13. Footprint Recovery
  14. Enhancement of 3D Reconstruction
  15. Qualitative Evaluation of Height Completion
  16. ...and 20 more sections

Figures (22)

  • Figure 1: RoofDiffusion restores height maps of challenging roof geometry, even under conditions of extreme sparsity, regional incompleteness, and noise. The bookend columns of point clouds are 3D views, the other columns are top views of height maps.
  • Figure 1: Distribution of height differences and roof counts.
  • Figure 2: Types of corrupted roof height maps with real-world scan.
  • Figure 2: Examples of tree height map in PoznanRD dataset. (The examples are color mapped and resized for clear visualization. The height increases as the color becomes more yellow.)
  • Figure 3: RoofDiffusion reconstruction of height maps corrupted by scan line strip patterns.
  • ...and 17 more figures