Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TreeON: Reconstructing 3D Tree Point Clouds from Orthophotos and Heightmaps

Angeliki Grammatikaki, Johannes Eschner, Pedro Hermosilla, Oscar Argudo, Manuela Waldner

TL;DR

This work introduces a new training supervision strategy that combines both geometric supervision and differentiable shadow and silhouette losses to learn point cloud representations of trees without requiring species labels, procedural rules, terrestrial reconstruction data, or ground laser scans.

Abstract

We present TreeON, a novel neural-based framework for reconstructing detailed 3D tree point clouds from sparse top-down geodata, using only a single orthophoto and its corresponding Digital Surface Model (DSM). Our method introduces a new training supervision strategy that combines both geometric supervision and differentiable shadow and silhouette losses to learn point cloud representations of trees without requiring species labels, procedural rules, terrestrial reconstruction data, or ground laser scans. To address the lack of ground truth data, we generate a synthetic dataset of point clouds from procedurally modeled trees and train our network on it. Quantitative and qualitative experiments demonstrate better reconstruction quality and coverage compared to existing methods, as well as strong generalization to real-world data, producing visually appealing and structurally plausible tree point cloud representations suitable for integration into interactive digital 3D maps. The codebase, synthetic dataset, and pretrained model are publicly available at https://angelikigram.github.io/treeON/.

TreeON: Reconstructing 3D Tree Point Clouds from Orthophotos and Heightmaps

TL;DR

This work introduces a new training supervision strategy that combines both geometric supervision and differentiable shadow and silhouette losses to learn point cloud representations of trees without requiring species labels, procedural rules, terrestrial reconstruction data, or ground laser scans.

Abstract

We present TreeON, a novel neural-based framework for reconstructing detailed 3D tree point clouds from sparse top-down geodata, using only a single orthophoto and its corresponding Digital Surface Model (DSM). Our method introduces a new training supervision strategy that combines both geometric supervision and differentiable shadow and silhouette losses to learn point cloud representations of trees without requiring species labels, procedural rules, terrestrial reconstruction data, or ground laser scans. To address the lack of ground truth data, we generate a synthetic dataset of point clouds from procedurally modeled trees and train our network on it. Quantitative and qualitative experiments demonstrate better reconstruction quality and coverage compared to existing methods, as well as strong generalization to real-world data, producing visually appealing and structurally plausible tree point cloud representations suitable for integration into interactive digital 3D maps. The codebase, synthetic dataset, and pretrained model are publicly available at https://angelikigram.github.io/treeON/.
Paper Structure (19 sections, 5 equations, 6 figures, 7 tables)

This paper contains 19 sections, 5 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Overview
  4. Synthetic Training Dataset Generation Pipeline
  5. Network Architecture
  6. Supervision Strategy
  7. Postprocessing and Rendering
  8. Experiments
  9. Datasets
  10. Implementation Details
  11. Evaluation Protocol
  12. Results
  13. Ablation Study
  14. Results on Austrian Landmark Trees
  15. Comparison to State-of-the-Art Models
  16. ...and 4 more sections

Figures (6)

  • Figure 1: Our neural-based framework, TreeON, reconstructs coherent tree point clouds for trees in real datasets using only an orthophoto and a Digital Surface Model (DSM).
  • Figure 2: Tree generation pipeline starting from (a) procedural mesh: sampled colored point cloud (b), artificial DSM point cloud (visualized as surface with interior points) (c) and orthophoto (d).
  • Figure 3: Orthophoto and DSM generation from a top-down orthographic camera. Orthophotos include trees and shadows under directional lighting, while DSMs DSM (visualized as a surface for illustration) are derived from per-pixel depth. Sun positions vary in azimuth and elevation.
  • Figure 4: Overview of the network architecture and training supervision (green).
  • Figure 5: (a) Ground-truth tree, (b) reconstruction without refinement, and (c) reconstruction with refinement.
  • ...and 1 more figures