This project is a collection of notebooks performing different methods to create graphs and semantic labels from the shapefiles data.
The most known geomaps open source is OSM, so the demo notebooks are adapted to use OSM shapefiles as can be found here: https://www.openstreetmap.org/
We also use IGN data, namely the BDTOPO®, available for research purposes under a license "Recherche et Enseignement": https ://geoservices.ign.fr
An example of the OSM data used in the project is available here: https://drive.google.com/file/d/131ZEmiO2KCfE-5cLIPmxBtaeaAXEGK80/view?usp=sharing It is the slightly processed OSM data (we merged buildings which touch each other). The polygons for Moselle deparments are given for the reference as well in a shape file image_polygons_57.shp.
Please, check the notebook:
The notebook shows how to create RNG attributed graphs using the algorithm introduced in [1].
The notebook shows how to create RNG attributed graphs using the Delaunay triangulation (check the difference between RNG and Delaunay graphs in [1])
We also left an old demo for the graph creation form the IGN data, but we don't provide the BD topo used along with the project. Please, check the notebook for details: data_parser_distinguished_graphs.ipynb
A project to get the .png maps of given resolution from shapefiles using Geopandas and Shapely, the notebook is available here:
Having an image as a jp2 file and labels as vector data in .shp format, extract the vector data using the bounding boxes from an image, and then save the result as a .png file.
The data can be downloaded from here: https://drive.google.com/drive/folders/12cWiUhE278O4y2Z-hV6rMROvnVUoqxP5?usp=sharing These graph datasets are derived from exceirpts of the reference vector database produced by the French Mapping Agency (https://www.ign.fr/), namely the BDTOPO® and Open Street Map (https://www.openstreetmap.org/), namely OSM.
Each dataset represents a French administrative unit (department), at a given valid time. The data are provided in ESRI shapefile format. They are represented using the Lambert 93 coordinates reference system (EPSG:2154). Each graph represents the geographic entites located in a bouding box of 200 m on each side, centered on a given point of interest, and their spatial relationships. The graphs are built for BBOX located the following French admnistratives units:
- Moselle
- Meurthe-et-Moselle
- Bas-Rhin
- Cotes d'Armor
A distinct graph is built for each POI-centered BBOX, and for the following years:
- 2004
- 2010
- 2014
- 2019
This is cross-temporal and cross-source dataset for 4 French departments. We separate the dataset to two main parts:
- Single source data (IGN data only, across time) -Cross-source data (OSM data and IGN data, a subset of 2 departments, OSM data are dated 2020)
Further, we provide the data in the form which is the most convenient for our two set of experiements, as described in [1]. There are 2 folders, each per_department, and the graphs IDs match each other across all the subfolders in one folder, but not across the higher folders in the hierarchy.
Each separate data IGN+'Year' has the following components (inspired by the other commong graph dataset structure):
IGN+'Year'_A.txt - edge list
IGN+'Year'_edge_labels.txt - edge labels
IGN+'Year'_graph_indicator.txt - list of the node-graph correspondences
IGN+'Year'_graph_labels.txt - graph areas IDs
IGN+'Year'_node_attributes.txt - continious node attributes
IGN+'Year'_node_labels.txt - discreet node labels
IGN+'Year'_node_positions.txt - center node coordinates
Each deparment contains 4 distinct dates for IGN data. OSM data we use are dated 2020, however, there are also 4 sets of data to match the cross-time cross-search data retrieval.
/2004
/2010
/2014
/2019
Please, check the scripts folder delievered with the datast.
To ease the use of our dataset, we provide the daloaders compatible with Pytorch Geometric (the code here is the modified code from Pytorch Geometric data loader)
You can find the daloader in the file:
ign_dataset.py
To upload the dataset, please import the dataloader from .py file to your workspace, then call: ign = IGNDataset('/path/to/the folder/with/txt', prefix ='IGN_19',use_node_attr=True) When loaded, the dataset can be passed to the torch_geometric.data.DataLoader as an argument.
You can also display view the graphs, but we use a different dataloader for that. Please, check the following jupyter notebook from scripts folder for demos and details: results_visualization_notebook.ipynb
More information on the graph creation process in provided in the article.
These data are made available for research purposes only, especially to test the images and graph matching/indexing approaches developed as part on the ALEGORIA project (ANR-17-CE38-0014-01), in particular in the work presented in the following article:
-
M. Khokhlova, N. Abadie, V. Gouet-Brunet and L. Chen, Learning embeddings for cross-time geographic areas represented as graphs, In the 36th ACM/SIGAPP Symposium On Applied Computing (SAC 2021) - Technical Track Geographic Information Analysis, 22-26 March 2021, Gwangju, South Korea.
Using them implies:
- complying with the terms of the following license agreement: https://geoservices.ign.fr/ressources_documentaires/Espace_documentaire/licence_gratuite.pdf
- citing the article mentioned above.
If you have any questions, please contact:
- about the ALEGORIA project: http://alegoria.ign.fr/en/contact/contact_us
- about terms of use of IGN data: contact@geoservices.ign.fr
We envisaged the final task as the similarity-based graph matching across years and databases.
However,the task is not always trivial, the resulting graphs can be pretty different for different databases and across years.
Full project requirements are specified in requirements.txt The minimal requirements for graph creation notebooks are:
- Networkx
- Geopandas
- Pandas
- Shapely
- Matplotlib
- Earthpy
- Pickle
All code found in this repository is licensed under GPLv3:
Copyright 2019-2020 Margarita Khokhlova, margarita.khokhlova@ign.fr, margokhokhlova@gmail.com
Graph creation Examples is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
Graph creation Examples are distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details https://www.gnu.org/licenses/.
This work is supported by ANR, the French National Re-search Agency, within the ALEGORIA project, under GrantANR-17-CE38-0014-01.
- Toussaint, G. T. (1980), "The relative neighborhood graph of a finite planar set", Pattern Recognition, 12 (4): 261–268, doi:10.1016/0031-3203(80)90066-7.
- Khokhlova, Margarita, Gouet-Brunet, V., Abadie, N., & Chen, L. "Cross-year multi-modal image retrieval using siamese networks." 2020 IEEE International Conference on Image Processing (ICIP). IEEE, 2020
- Khokhlova, Margarita, Gouet-Brunet, V., Abadie, N., & Chen, L. "Recherche multimodale d'images aériennes multi-date à l'aide d'un réseau siamois." RFIAP. 2020.