ipoints = np.argmin(distance_matrix.euclidian(points, rpoints), axis=1)
npoints = len(ipoints)
# geodesic distances
rE = distance_matrix.euclidian(rpoints)
rC = delaunay_2d.connectivity(rpoints)
rG, rparents = fastm.fast_marching_3d_aniso(
# im*np.mean(rE[rC==1]),
imw,
rpoints,
heap_size=1e6,
offset=1e-5,
connectivity26=True,
output_arguments=('DM', 'parents'),
)
pathes, connections = fastm.shortestpath(rparents)
impath = np.zeros(imw.shape)
for path in pathes:
impath[tuple(pathes[path].T)] = path
if 0:
# rM = rE*(rG<1e9)/(rG*(rG<1e9) + (rG>1e9))
# rM = (rG >= 3*rE)
rM = np.minimum(np.exp((rG - 3 * rE) / 10), 1)
# Mcom = commute_time(1-rM)[ipoints,:][:,ipoints]
# Mcom = Mcom - np.min(Mcom[np.where(np.triu(np.ones(Mcom.shape),k=1))])
# Mcom.flat[::Mcom.shape[0]+1] = 0
# Mcom /= np.c_[np.sum(Mcom, axis=1)]
Aff = np.exp(-rG**2 / np.mean(rG[rG < 1e9] * 3)**2)
points = np.c_[
np.random.randint(0, shape[0], npoints),
np.random.randint(0, shape[1], npoints),
]
dist, labels, DM, inter_ = fastm.fast_marching_3d(
speed,
points,
offset=1,
output_arguments=(
'distances',
'labels',
'DM',
'intersections',
)
)
inter = np.argwhere(inter_!=-1)
path = fastm.shortestpath(points, inter_, dist)
path_ = np.argwhere(path!=-1)
col = path[tuple(path_.T)]
if 1:
from matplotlib import pyplot
pyplot.figure()
pyplot.imshow(speed.T, cmap='gray', aspect='equal', interpolation='nearest')
pyplot.imshow(labels.T, alpha=0.8, interpolation='nearest')
pyplot.scatter(points[:,0], points[:,1])
pyplot.scatter(inter[:,0], inter[:,1], c='r')
pyplot.scatter(path_[:,0], path_[:,1], c=col)
pyplot.show()
# geodesic distances
rE = distance_matrix.euclidian(rpoints)
rC = delaunay_2d.connectivity(rpoints)
rG,rparents = fastm.fast_marching_3d_aniso(
# im*np.mean(rE[rC==1]),
imw,
rpoints,
heap_size=1e6,
offset=1e-5,
connectivity26=True,
output_arguments=('DM','parents'),
)
pathes,connections = fastm.shortestpath(rparents)
impath = np.zeros(imw.shape)
for path in pathes:
impath[tuple(pathes[path].T)] = path
if 0:
# rM = rE*(rG<1e9)/(rG*(rG<1e9) + (rG>1e9))
# rM = (rG >= 3*rE)
rM = np.minimum(np.exp((rG - 3*rE)/10),1)
# Mcom = commute_time(1-rM)[ipoints,:][:,ipoints]
# Mcom = Mcom - np.min(Mcom[np.where(np.triu(np.ones(Mcom.shape),k=1))])
# Mcom.flat[::Mcom.shape[0]+1] = 0
# Mcom /= np.c_[np.sum(Mcom, axis=1)]
Aff = np.exp(-rG**2/np.mean(rG[rG<1e9]*3)**2)