def test_stitch_conf(self):
src_im = skimage.io.imread(
os.path.join(data_dir, 'sample_fp_mask_from_geojson.tif'))
src_im[src_im != 0] = 1
src_im = src_im.astype('float64')
inf_tiler = sol.nets.datagen.InferenceTiler('keras',
width=250,
height=250)
tiles, tile_inds, _ = inf_tiler(src_im)
rands = np.array([
0.93794284, 0.88778908, 0.25066594, 0.76800494, 0.43465608,
0.69903218, 0.13256956, 0.20246324, 0.65134984, 0.98667763,
0.38168734, 0.85653983, 0.34337332, 0.75118759, 0.01128917,
0.92725672
])
# scale the values for heterogeneity
tiles = tiles * rands[:, np.newaxis, np.newaxis, np.newaxis]
result = stitch_images(tiles,
idx_refs=tile_inds,
out_width=900,
out_height=900,
method='confidence')
expected_result = np.load(
os.path.join(data_dir, 'stitching_conf_output.npy'))
assert np.array_equal(result, expected_result)
def test_stitch_InferenceTiler_output(self):
inf_tiler = sol.nets.datagen.InferenceTiler('keras',
width=250,
height=250)
tiles, tile_inds, _ = inf_tiler(
os.path.join(data_dir, 'sample_geotiff.tif'))
restored_im = stitch_images(tiles,
idx_refs=tile_inds,
out_width=900,
out_height=900)
expected_result = sol.utils.io.imread(
os.path.join(data_dir, 'sample_geotiff.tif'))
assert np.array_equal(restored_im[:, :, 0], expected_result)
def test_stitch_firstval(self):
inf_tiler = sol.nets.datagen.InferenceTiler('keras',
width=250,
height=250)
tiles, tile_inds, _ = inf_tiler(
os.path.join(data_dir, 'sample_geotiff.tif'))
tiles[11, :, :, :] = tiles[11, :, :, :] + 10 # to have a diff to check
result = stitch_images(tiles,
idx_refs=tile_inds,
out_width=900,
out_height=900,
method='first')
expected_result = np.load(
os.path.join(data_dir, 'stitching_first_output.npy'))
assert np.array_equal(result, expected_result)