Utilities for modeling and visualizing SpaceNet data.
SpaceNet data is available on aws.
spacenet-utils dependencies: numpy, pandas, osgeo, matplotlib, geomet, scikit-learn.
Display geotiff image with building-polygon overlay.
# python3
import spacenet_utils as snu
import matplotlib.pyplot as plt
# path to csv of ground-truth polygon vertices
csv_path = 'AOI_5_Khartoum_Train/summaryData/AOI_5_Khartoum_Train_Building_Solutions.csv'
# path to geotiff image
im_path = 'AOI_5_Khartoum_Train/MUL-PanSharpen/MUL-PanSharpen_AOI_5_Khartoum_img5.tif'
fig, ax = plt.subplots()
snu.plot_image(ax,im_path,band=7)
snu.plot_gt(ax,im_path,csv_path)
plt.show()
Label buildings using kNN on the pixel spectra.
$ python example_knn.py
performance = 0.799041
confusion matrix =
array([[288411, 13766],
[ 71139, 49184]])
fscore = 0.536729360026example_knn.py has only been tested on MUL-PanSharpened images from the Khartoum region.
To use on a different region or image type, be sure to set the following kwargs in spacenet_utils.py: poly_verts2mask.w, postprocess.w, get_poly_arr.im_id_prefix, plot_gt.im_id_prefix.
See these functions' docstrings for further documentation.