def test_sphere_distance():
# make a circle, go around...
radius = 3.2
n = 5000
n2 = n / 2
# pi at point n2 in array
angles = np.linspace(0, np.pi*2, n, endpoint=False)
x = np.sin(angles) * radius
y = np.cos(angles) * radius
# dists around half circle, including pi
half_x = x[:n2+1]
half_y = y[:n2+1]
half_dists = np.sqrt(np.diff(half_x)**2 + np.diff(half_y)**2)
# approximate distances from 0 to pi (not including 0)
csums = np.cumsum(half_dists)
# concatenated with distances from pi to 0 again
cdists = np.r_[0, csums, csums[-2::-1]]
# check approximation close to calculated
sph_d = sphere_distance([0, radius], np.c_[x, y])
yield assert_array_almost_equal, cdists, sph_d
# Now check with passed radius
sph_d = sphere_distance([0, radius], np.c_[x, y], radius=radius)
yield assert_array_almost_equal, cdists, sph_d
# Check points not on surface raises error when asked for
yield assert_raises, ValueError, sphere_distance, [1, 0], [0, 2]
# Not when check is disabled
sph_d = sphere_distance([1, 0], [0, 2], None, False)
# Error when radii don't match passed radius
yield assert_raises, ValueError, sphere_distance, [1, 0], [0, 1], 2.0
def direction_to_classification(sphere,
next_dir,
include_stop=False,
last_is_stop=False,
stop_values=None):
# code adapted from Benou "DeepTract",exi
# https://github.com/itaybenou/DeepTract/blob/master/utils/train_utils.py
sl_len = len(next_dir)
loop_len = sl_len - 1 if include_stop and last_is_stop else sl_len
l = len(sphere.theta) + 1 if include_stop else len(sphere.theta)
classification_output = np.zeros((sl_len, l))
for i in range(loop_len):
if not (next_dir[i, 0] == 0.0 and next_dir[i, 1] == 0.0
and next_dir[i, 2] == 0.0):
labels_odf = np.exp(
-1 *
sphere_distance(next_dir[i, :],
np.asarray([sphere.x, sphere.y, sphere.z]).T,
radius=1,
check_radius=False) * 10)
if include_stop:
classification_output[i][:-1] = labels_odf / np.sum(labels_odf)
classification_output[i, -1] = 0.0
else:
classification_output[i] = labels_odf / np.sum(labels_odf)
if include_stop and last_is_stop:
classification_output[-1, -1] = 1 # stop condition or
if include_stop and stop_values is not None:
classification_output[:, -1] = stop_values # stop values
return classification_output
def find_viewpoint_id(sphere_points, point):
# print(np.array(point))
# distances = np.linalg.norm(sphere_points - point, axis=1)
distances = sphere_distance(sphere_points, np.array(point), radius=1)
viewpoint_index = np.argmin(distances)
# print(distances[viewpoint_index])
return viewpoint_index
def smooth_labels(directions, Num_outputs=725):
smoothed_labels = np.zeros((directions.shape[0], directions.shape[1], Num_outputs), np.float32)
sphere_points = get_sphere('repulsion724')
for i in range(directions.shape[0]):
aux_bool = 0
for j in range(directions.shape[1]):
if not (directions[i, j, 0] == 0.0 and directions[i, j, 0] == 0.0 and directions[i, j, 0] == 0.0):
idx = sphere_points.find_closest(directions[i, j, :])
labels_odf = np.exp(-1 * sphere_distance(directions[i, j, :], np.asarray(
[sphere_points.x, sphere_points.y, sphere_points.z]).T, radius=1.0) / 0.1)
labels_odf = labels_odf / np.sum(labels_odf)
smoothed_labels[i, j, :-1] = labels_odf
smoothed_labels[i, j, -1] = 0.0
elif aux_bool == 0:
smoothed_labels[i, j, -1] = 1.0
aux_bool = 1
return smoothed_labels
