Module cycpd.affine_registration
Expand source code
import time
from builtins import super
import numpy as np
from .expectation_maximization_registration import expectation_maximization_registration
class affine_registration(expectation_maximization_registration):
"""
"""
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
tic = time.time()
self.B = np.eye(self.D)
self.t = np.zeros(self.D)
self.B1 = None
self.B2 = None
toc = time.time()
self.time_to_initialize_registration = toc - tic
def update_transform(self):
"""
Update the transform parameters.
"""
self.muX = np.divide(np.matmul(self.X.T, self.Pt1), self.Np)
self.muY = np.divide(
np.matmul(self.Y.T, self.P1), self.Np
) # changed previous code from PyCPD because we no longer store self.P
self.B1 = np.matmul(self.PX.T, self.Y) - self.Np * np.matmul(
self.muX[:, None], self.muY[:, None].T
)
self.B2 = np.matmul((self.Y * self.P1[:, None]).T, self.Y) - self.Np * np.matmul(
self.muY[:, None], self.muY[:, None].T
)
self.B = np.linalg.solve(self.B2.T, self.B1.T).T
self.t = np.squeeze(self.muX[:, None] - np.matmul(self.B, self.muY[:, None]))
self.err = abs(self.E - self.E_old)
def transform_point_cloud(self, Y=None):
"""
Transform a point cloud.
Parameters
----------
Y : array_like, optional
The point cloud to transform.
"""
if Y is None:
self.TY = np.dot(self.Y, self.B.T) + self.t
return
else:
return np.dot(Y, self.B.T) + self.t
def update_variance(self):
"""
Update the variance.
"""
self.sigma2_prev = self.sigma2
A = np.sum(np.square(self.X) * self.Pt1[:, None])
B = self.Np * np.matmul(self.muX[:, None].T, self.muX[:, None])
C = np.trace(np.matmul(self.B1, self.B.T))
self.sigma2 = abs(A - B - C) / (self.Np * self.D)
if self.sigma2 <= 0:
self.sigma2 = self.tolerance / 10
def get_registration_parameters(self):
"""
Get the registration parameters.
Returns
-------
self.B : array_like
The affine transform matrix.
self.t : array_like
The translation vector.
"""
return self.B, self.tClasses
class affine_registration (*args, **kwargs)-
Expand source code
class affine_registration(expectation_maximization_registration): """ """ def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) tic = time.time() self.B = np.eye(self.D) self.t = np.zeros(self.D) self.B1 = None self.B2 = None toc = time.time() self.time_to_initialize_registration = toc - tic def update_transform(self): """ Update the transform parameters. """ self.muX = np.divide(np.matmul(self.X.T, self.Pt1), self.Np) self.muY = np.divide( np.matmul(self.Y.T, self.P1), self.Np ) # changed previous code from PyCPD because we no longer store self.P self.B1 = np.matmul(self.PX.T, self.Y) - self.Np * np.matmul( self.muX[:, None], self.muY[:, None].T ) self.B2 = np.matmul((self.Y * self.P1[:, None]).T, self.Y) - self.Np * np.matmul( self.muY[:, None], self.muY[:, None].T ) self.B = np.linalg.solve(self.B2.T, self.B1.T).T self.t = np.squeeze(self.muX[:, None] - np.matmul(self.B, self.muY[:, None])) self.err = abs(self.E - self.E_old) def transform_point_cloud(self, Y=None): """ Transform a point cloud. Parameters ---------- Y : array_like, optional The point cloud to transform. """ if Y is None: self.TY = np.dot(self.Y, self.B.T) + self.t return else: return np.dot(Y, self.B.T) + self.t def update_variance(self): """ Update the variance. """ self.sigma2_prev = self.sigma2 A = np.sum(np.square(self.X) * self.Pt1[:, None]) B = self.Np * np.matmul(self.muX[:, None].T, self.muX[:, None]) C = np.trace(np.matmul(self.B1, self.B.T)) self.sigma2 = abs(A - B - C) / (self.Np * self.D) if self.sigma2 <= 0: self.sigma2 = self.tolerance / 10 def get_registration_parameters(self): """ Get the registration parameters. Returns ------- self.B : array_like The affine transform matrix. self.t : array_like The translation vector. """ return self.B, self.tAncestors
Methods
def get_registration_parameters(self)-
Get the registration parameters.
Returns
self.B : array_like- The affine transform matrix.
self.t : array_like- The translation vector.
Expand source code
def get_registration_parameters(self): """ Get the registration parameters. Returns ------- self.B : array_like The affine transform matrix. self.t : array_like The translation vector. """ return self.B, self.t def transform_point_cloud(self, Y=None)-
Transform a point cloud.
Parameters
Y:array_like, optional- The point cloud to transform.
Expand source code
def transform_point_cloud(self, Y=None): """ Transform a point cloud. Parameters ---------- Y : array_like, optional The point cloud to transform. """ if Y is None: self.TY = np.dot(self.Y, self.B.T) + self.t return else: return np.dot(Y, self.B.T) + self.t def update_transform(self)-
Update the transform parameters.
Expand source code
def update_transform(self): """ Update the transform parameters. """ self.muX = np.divide(np.matmul(self.X.T, self.Pt1), self.Np) self.muY = np.divide( np.matmul(self.Y.T, self.P1), self.Np ) # changed previous code from PyCPD because we no longer store self.P self.B1 = np.matmul(self.PX.T, self.Y) - self.Np * np.matmul( self.muX[:, None], self.muY[:, None].T ) self.B2 = np.matmul((self.Y * self.P1[:, None]).T, self.Y) - self.Np * np.matmul( self.muY[:, None], self.muY[:, None].T ) self.B = np.linalg.solve(self.B2.T, self.B1.T).T self.t = np.squeeze(self.muX[:, None] - np.matmul(self.B, self.muY[:, None])) self.err = abs(self.E - self.E_old) def update_variance(self)-
Update the variance.
Expand source code
def update_variance(self): """ Update the variance. """ self.sigma2_prev = self.sigma2 A = np.sum(np.square(self.X) * self.Pt1[:, None]) B = self.Np * np.matmul(self.muX[:, None].T, self.muX[:, None]) C = np.trace(np.matmul(self.B1, self.B.T)) self.sigma2 = abs(A - B - C) / (self.Np * self.D) if self.sigma2 <= 0: self.sigma2 = self.tolerance / 10
Inherited members