buzzardpython/ci repository overview

import buzzard as buzz
import numpy as np
import matplotlib.pyplot as plt

# Open the files. Only files' metadata are read so far
r = buzz.open_raster('path/to/rgba-image.tif')
v = buzz.open_vector('path/to/polygons.geojson', driver='GeoJSON')


# Load the polygons from disk one by one as shapely objects
for poly in v.iter_data():

    # Compute the Footprint bounding `poly`
    fp = r.fp.intersection(poly)
    print(fp)

    # Load the image from disk at `fp` to a numpy array
    rgb = r.get_data(fp=fp, channels=(0, 1, 2))
    alpha = r.get_data(fp=fp, channels=3)

    # Create a boolean mask as a numpy array from the shapely polygon
    mask = np.invert(fp.burn_polygons(poly))

    # Darken pixels outside of polygon, set transparent pixels to orange
    rgb[mask] = (rgb[mask] * 0.5).astype(np.uint8)
    rgb[alpha == 0] = [236, 120, 57]

    # Show the result with matplotlib
    plt.imshow(rgb)
    plt.show()

import buzzard as buzz
import numpy as np
import matplotlib.pyplot as plt
import keras

r = buzz.open_raster('path/to/rgba-image.tif')
km = keras.models.load_model('path/to/deep-learning-model.hdf5')

# Chunk the raster's Footprint to Footprints of size
# 1920 x 1080 pixel stored in a 2d numpy array
tiles = r.fp.tile(1920, 1080)

all_roads = []

for i, fp in enumerate(tiles.flat):
    rgb = r.get_data(fp=fp, channels=(0, 1, 2))

    # Perform pixelwise semantic segmentation with a keras model
    predictions_heatmap = km.predict(rgb[np.newaxis, ...])[0]
    predictions_top1 = np.argmax(predictions_heatmap, axis=-1)

    # Save the prediction to a `geotiff`
    with buzz.create_raster(path='predictions_{}.tif'.format(i), fp=fp,
                            dtype='uint8', channel_count=1).close as out:
        out.set_data(predictions_top1)

    # Extract the road polygons by transforming a numpy boolean mask to shapely polygons
    road_polygons = fp.find_polygons(predictions_top1 == 3)
    all_roads += road_polygons

    # Show the result with matplotlib for one tile
    if i == 2:
        plt.imshow(rgb)
        plt.imshow(predictions_top1)
        plt.show()

# Save all roads found to a single `shapefile`
with buzz.create_vector(path='roads.shp', type='polygon').close as out:
    for poly in all_roads:
        out.inser_data(poly)

# Step 1 - Install Anaconda
# https://www.anaconda.com/download/

# Step 2 - Create env
conda create -n buzz python gdal>=2.3.3 shapely rtree -c 'conda-forge'

# Step 3 - Activate env
conda activate buzz

# Step 4 - Install buzzard
pip install buzzard

docker build -t buzz --build-arg PYTHON_VERSION=3.7 https://raw.githubusercontent.com/earthcube-lab/buzzard/master/.circleci/images/base-python/Dockerfile
docker run -it --rm buzz bash
pip install buzzard

# Step 1 - Install GDAL and rtree ******************************************* **
# Windows
# https://www.lfd.uci.edu/~gohlke/pythonlibs/#gdal
# https://www.lfd.uci.edu/~gohlke/pythonlibs/#rtree

# MacOS
brew install gdal
brew tap osgeo/osgeo4mac
brew tap --repair
brew install gdal2
brew install spatialindex
export PATH="/usr/local/opt/gdal2/bin:$PATH"
python3 -m pip install 'gdal==2.3.3'

# Ubuntu
# Run the commands from the following Dockerfile:
# https://github.com/earthcube-lab/buzzard/blob/master/doc/ubuntu_install/Dockerfile

# Step 2 - Install buzzard ************************************************** **
python3 -m pip install buzzard

git clone https://github.com/earthcube-lab/buzzard
pip install -r buzzard/requirements-dev.txt
pytest buzzard/buzzard/test

cd docs
make html
open _build/html/index.html