Objectives : finding all the trees on public land in a city
categorizing kind of tree from street view data
start by loading marseille_tree_gis.qgz
using QGIS 3.6
found in dataset_marseille/deepforest_pred_*
polygons generated using deepforest and deepforest_marseille/run_predictions_geotiff.py
also see:
deepforest_marseille/explore predictions.ipynb
using Jupyter notebook
-
IGN Geoportail (best quality - limited access) https://geoservices.ign.fr/blog/2017/06/28/geoportail_sans_compte.html
WMTS url: https://wxs.ign.fr/pratique/geoportail/wmts?SERVICE=WMTS&REQUEST=GetCapabilities
-
Mapbox (ok quality) https://docs.mapbox.com/help/tutorials/mapbox-arcgis-qgis/
dataset_marseille/contours_quartiers_Marseille.shp
Découpage administratif des 111 quartiers de Marseille (par agrégation d'IRIS). https://www.data.gouv.fr/fr/datasets/quartiers-de-marseille/#resource-b1e544a4-f065-494e-8012-843c6cc63cfc
With QGIS 3.6 :
= Step1 : Do a geotiff raster export of map tiles =
- right click XY layer using * Satellite Raster Views * above
- save as : geotiff
- deselect "Create VRT"
- Extent : map canvas extent (just the window viewport for a sample)
- Resolution: 0.2 Horizontal and 0.2 Vertical, this means 0.2m per pixel. (this is the max that IGN supports).
- Raster file .tif is created! - and will be added to QGIS for visualization
= Step2 : deepforest =
- Run deepforest prediction
python deepforest_marseille/run_predictions_geotiff.py input_geotiff_raster.tif predictions.shp
XXX at the moment the .py is broken but see
deepforest_marseille/explore predictions.ipynb
using jupyter notebook for working code.
predictions.shp
can be added back withLayer > Add Layer > Add Vector Layer
the conda environment to be able to run deepforest is
exported as deepforest_conda_environment.yml
install it with
conda env create -f deepforest_conda_environment.yml