PyPVRoof: a Python package for extracting the metadata of rooftop PV installations from their polygon

This repository contains the necessary data and notebooks to replicate the results and analyses carried out in the paper "PyPVRoof: a Python package for the modular extraction of minimal metadata of rooftop PV installations".

The project repository is accessible here: https://github.com/gabrielkasmi/pypvroof

The paper is accessible here: 

This repository is self-sufficient to replicate the results. This repository also hosts the data necessary to run the notebook hands-on.ipynb from the public repository of the project.

The dataset is organized as follows:

• characteristics/ root folder for the replication of the results
  • /CSV: contains the `.csv` files necessary to train and evaluate the methods
  • /Data: main folder for the data
    • /BDAPPV: folder containing the data stemming from BDAPPV.
    • /DSFrance: folder containing auxiliary data to run the methods on the outputs of DeepPVMapper
    •  /LiDAR: folder containing a sample of LiDAR images for evaluation
    •  /Yann: folder containing auxiliary data and annotations. 
  • /models: folder containing the weights of the models
  • /notebooks: folder containing the Jupyter notebook in which the methods are evaluated.
    • Characteristics_Extraction_Methods.ipynb is the notebook in which the methods are gathered and defined.
    • The other notebooks can be used to generate the models using the data located in the /Data folder.
  • /scalers: folder containing the scaler for the random forest defined for tilt and azimuth estimation
  •  env.yml: use this file to defined a virtual environment from which the notebooks can be launched. 
  • README.md: the readme file
• hands-on/ Root folder to run the hands-on notebook from the repository, accessible here: : https://github.com/gabrielkasmi/pypvroof/blob/master/hands-on.ipynb
  • bdappv-metadata.csv: a file containing ground truth information for installations located in France. This dataset is filtered from the BDAPPV dataset. Localizations were added based on the information on the departement.
  • arrays_69.geojson. A raw file coming from DeepPVMapper' first step (detection and segmentation). This file is the output of the hands-on notebook of DeepPVMapper. You can access the repository here: https://github.com/gabrielkasmi/deeppvmapper
  • lookup-table.json. The lookup table of DeepPVMapper
  • tile.tif: a LiDAR raster as an example for the tilt and azimuth estimation using Theil-Sen algorithm