def test_03_01_shape_dense(self):
r = np.random.RandomState()
r.seed(101)
labels = r.randint(0, 10, size=(2, 3, 4, 5, 6, 7))
s = cpo.Segmentation(dense=labels)
self.assertTrue(s.has_shape())
self.assertEqual(tuple(s.shape), tuple(labels.shape[1:]))
def test_02_01_sparse_to_dense(self):
#
# Make 10 circles that might overlap
#
r = np.random.RandomState()
r.seed(201)
i, j = np.mgrid[0:50, 0:50]
ii = []
jj = []
vv = []
for idx in range(10):
x_loc = r.uniform() * 30 + 10
y_loc = r.uniform() * 30 + 10
max_radius = np.min(
[min(loc - 1, 49 - loc) for loc in x_loc, y_loc])
radius = r.uniform() * (max_radius - 5) + 5
mask = ((i - y_loc)**2 + (j - x_loc)**2) <= radius**2
ii.append(i[mask])
jj.append(j[mask])
vv.append(np.ones(np.sum(mask), np.uint32) * (idx + 1))
ijv = np.core.records.fromarrays(
[np.hstack(x) for x in ii, jj, vv],
[(HDF5ObjectSet.AXIS_Y, np.uint32, 1),
(HDF5ObjectSet.AXIS_X, np.uint32, 1),
(HDF5ObjectSet.AXIS_LABELS, np.uint32, 1)])
s = cpo.Segmentation(sparse=ijv, shape=(1, 1, 1, 50, 50))
dense, indices = s.get_dense()
self.assertEqual(tuple(dense.shape[1:]), (1, 1, 1, 50, 50))
self.assertEqual(np.sum(dense > 0), len(ijv))
retrieval = dense[:, 0, 0, 0, ijv[HDF5ObjectSet.AXIS_Y],
ijv[HDF5ObjectSet.AXIS_X]]
matches = (retrieval == ijv[HDF5ObjectSet.AXIS_LABELS][None, :])
self.assertTrue(np.all(np.sum(matches, 0) == 1))
def test_01_01_dense(self):
r = np.random.RandomState()
r.seed(101)
labels = r.randint(0, 10, size=(2, 3, 4, 5, 6, 7))
s = cpo.Segmentation(dense=labels)
self.assertTrue(s.has_dense())
self.assertFalse(s.has_sparse())
np.testing.assert_array_equal(s.get_dense()[0], labels)
def test_03_02_shape_sparse_explicit(self):
r = np.random.RandomState()
r.seed(102)
shape = (1, 1, 1, 50, 50)
ijv = np.core.records.fromarrays(
[r.randint(0, 10, size=20) for _ in range(3)],
[(HDF5ObjectSet.AXIS_Y, np.uint32, 1),
(HDF5ObjectSet.AXIS_X, np.uint32, 1),
(HDF5ObjectSet.AXIS_LABELS, np.uint32, 1)])
s = cpo.Segmentation(sparse=ijv, shape=shape)
self.assertTrue(s.has_shape())
self.assertEqual(tuple(s.shape), shape)
def test_01_02_sparse(self):
r = np.random.RandomState()
r.seed(102)
ijv = np.core.records.fromarrays(
[r.randint(0, 10, size=20) for _ in range(3)],
[(HDF5ObjectSet.AXIS_Y, np.uint32, 1),
(HDF5ObjectSet.AXIS_X, np.uint32, 1),
(HDF5ObjectSet.AXIS_LABELS, np.uint32, 1)])
s = cpo.Segmentation(sparse=ijv)
np.testing.assert_array_equal(s.sparse, ijv)
self.assertFalse(s.has_dense())
self.assertTrue(s.has_sparse())
def test_02_02_dense_to_sparse(self):
#
# Make 10 circles that might overlap
#
r = np.random.RandomState()
r.seed(201)
i, j = np.mgrid[0:50, 0:50]
dense = np.zeros((10, 1, 1, 1, 50, 50), np.uint32)
for idx in range(10):
x_loc = r.uniform() * 30 + 10
y_loc = r.uniform() * 30 + 10
max_radius = np.min([min(loc - 1, 49 - loc) for loc in x_loc, y_loc])
radius = r.uniform() * (max_radius - 5) + 5
mask = ((i - y_loc) ** 2 + (j - x_loc) ** 2) <= radius ** 2
dense[idx, 0, 0, 0, mask] = idx + 1
s = cpo.Segmentation(dense=dense)
ijv = s.sparse
self.assertEqual(np.sum(dense > 0), len(ijv))
retrieval = dense[:, 0, 0, 0,
ijv[HDF5ObjectSet.AXIS_Y], ijv[HDF5ObjectSet.AXIS_X]]
matches = (retrieval == ijv[HDF5ObjectSet.AXIS_LABELS][None, :])
self.assertTrue(np.all(np.sum(matches, 0) == 1))
