def test_build_full_labelled_image(tanzania_image_size, tanzania_nb_labels,
tanzania_raw_image_size):
"""Test the label prediction on a high-resolution image that has to be
tiled during inference process
The labelled output is composed of label IDs that must corresponds to the
dataset glossary, and its shape must equal the original image size.
"""
datapath = "./tests/data"
dataset = "tanzania"
image_paths = postprocess.get_image_paths(datapath, dataset,
tanzania_image_size,
"tanzania_sample")
images = postprocess.extract_images(image_paths)
coordinates = postprocess.extract_coordinates_from_filenames(image_paths)
model = postprocess.get_trained_model(datapath, dataset,
tanzania_image_size,
tanzania_nb_labels)
labelled_image = postprocess.build_full_labelled_image(
images,
coordinates,
model,
tile_size=tanzania_image_size,
img_width=tanzania_raw_image_size,
batch_size=2,
)
assert labelled_image.shape == (
tanzania_raw_image_size,
tanzania_raw_image_size,
)
assert np.all(
[ul in range(tanzania_nb_labels) for ul in np.unique(labelled_image)])
def test_extract_coordinates_from_filenames():
"""Test the x, y coordinates extraction from filenames
The filename chosen convention is as follows:
<path>/<name>_<width>_<height>_<x>_<y>.<extension>
"""
filenames = [
"./foofolder/foo_100_100_0_1000.png",
"./barfolder/bar_100_100_2000_2100.png",
]
coordinates = postprocess.extract_coordinates_from_filenames(filenames)
assert np.all(coordinates == [[0, 1000], [2000, 2100]])