Synthetic learning for primitive-based building model reconstruction from point clouds

This repository contains code for the paper "Synthetic learning for primitive-based building model reconstruction from point clouds". The project is implemented in Python and PyTorch, with additional utilities for evaluation and visualization.

Roof Primitives

Design of the roof primitives, including the parameters and visualizations.

Roof Primitives

Design of the training procedure.

Features

• Point Cloud Processing: Implements point cloud sampling, grouping, and feature extraction using neural networks.
• Roof Primitive Determination: Predicts roof types and geometric parameters of roof primitives from point cloud data.
• Visualization: Uses pyvista for 3D visualization of predictions and ground truth.
• Evaluation Metrics: Includes metrics like RMSE, reconstruction scores, and parameter errors for model evaluation.

Repository Structure

• demo: Demo scripts for building roof primitive determination from building point clouds.
• source: Scripts the work in this repo, including model design, training, testing, and other utilities.
• main_**: Maing entrance for building primitive determination.

Reconstruction Results

Requirements

• Python 3.8+
• PyTorch 1.10+
• NumPy
• PyVista
• Matplotlib
• Pandas
• Scipy

Install the required packages using:

pip install -r requirements.txt

Citations

@ARTICLE{Li2019-iz,
  title   = "{GEOMETRIC} {OBJECT} {BASED} {BUILDING} {RECONSTRUCTION} {FROM}
             {SATELLITE} {IMAGERY} {DERIVED} {POINT} {CLOUDS}",
  author  = "Li, Zhixin and Xu, Bo and Shan, Jie",
  journal = "International Archives of the Photogrammetry, Remote Sensing \&
             Spatial Information Sciences",
  year    =  2019
}

@ARTICLE{Zhang2021-as,
  title     = "Optimal model fitting for building reconstruction from point
               clouds",
  author    = "Zhang, Wenyuan and Li, Zhixin and Shan, Jie",
  journal   = "IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.",
  publisher = "Institute of Electrical and Electronics Engineers (IEEE)",
  volume    =  14,
  pages     = "9636--9650",
  year      =  2021
}

@ARTICLE{Li2022-zb,
  title     = "{RANSAC}-based multi primitive building reconstruction from {3D}
               point clouds",
  author    = "Li, Zhixin and Shan, Jie",
  journal   = "ISPRS J. Photogramm. Remote Sens.",
  publisher = "Elsevier BV",
  month     =  jan,
  year      =  2022,
  language  = "en"
}