mtourne/marseille_trees

finding all trees in marseille - work in progress & datasets

Tree challenge Marseille

Objectives : finding all the trees on public land in a city

categorizing kind of tree from street view data

QGIS

start by loading marseille_tree_gis.qgz using QGIS 3.6

Dataset

Predictions

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

Satellite Raster views

• 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/

  WMTS url: https://api.mapbox.com/styles/v1/mattiou/ckkl0f8zq3bzd17mqktg9mwqq/wmts?access_token=pk.eyJ1IjoibWF0dGlvdSIsImEiOiJja2trZWN6M2ExN2VqMnVwZ3liaTN0N2tjIn0.comgaQnS2e6Y6a3L_yHFfw

Contours Quartiers

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

deepforest predictions workflow

With QGIS 3.6 :

= Step1 : Do a geotiff raster export of map tiles =

1. right click XY layer using * Satellite Raster Views * above
2. save as : geotiff
3. deselect "Create VRT"
4. Extent : map canvas extent (just the window viewport for a sample)
5. Resolution: 0.2 Horizontal and 0.2 Vertical, this means 0.2m per pixel. (this is the max that IGN supports).
6. Raster file .tif is created! - and will be added to QGIS for visualization

= Step2 : deepforest =

7. 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.

8. predictions.shp can be added back with Layer > Add Layer > Add Vector Layer

deepforest install guide

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