def test_delete_cache(self):
x, y = util.make_grid(numPix=20, deltapix=1.)
gauss = Gaussian()
flux = gauss.function(x, y, amp=1., center_x=0., center_y=0., sigma=1.)
image = util.array2image(flux)
interp = Interpol()
kwargs_interp = {
'image': image,
'scale': 1.,
'phi_G': 0.,
'center_x': 0.,
'center_y': 0.
}
output = interp.function(x, y, **kwargs_interp)
assert hasattr(interp, '_image_interp')
interp.delete_cache()
assert not hasattr(interp, '_image_interp')
class SLIT_Starlets(object):
"""
Decomposition of an image using the Isotropic Undecimated Walevet Transform,
also known as "starlet" or "B-spline", using the 'a trous' algorithm.
Astronomical data (galaxies, stars, ...) are often very sparsely represented in the starlet basis.
Based on Starck et al. : https://ui.adsabs.harvard.edu/abs/2007ITIP...16..297S/abstract
"""
param_names = [
'amp', 'n_scales', 'n_pixels', 'scale', 'center_x', 'center_y'
]
lower_limit_default = {
'amp': [0],
'n_scales': 2,
'n_pixels': 5,
'center_x': -1000,
'center_y': -1000,
'scale': 0.000000001
}
upper_limit_default = {
'amp': [1e8],
'n_scales': 20,
'n_pixels': 1e10,
'center_x': 1000,
'center_y': 1000,
'scale': 10000000000
}
def __init__(self,
thread_count=1,
fast_inverse=True,
second_gen=False,
show_pysap_plots=False,
force_no_pysap=False):
"""
Load pySAP package if found, and initialize the Starlet transform.
:param thread_count: number of threads used for pySAP computations
:param fast_inverse: if True, reconstruction is simply the sum of each scale (only for 1st generation starlet transform)
:param second_gen: if True, uses the second generation of starlet transform
:param show_pysap_plots: if True, displays pySAP plots when calling the decomposition method
:param force_no_pysap: if True, does not load pySAP and computes starlet transforms in python.
"""
self.use_pysap, pysap = self._load_pysap(force_no_pysap)
if self.use_pysap:
self._transf_class = pysap.load_transform(
'BsplineWaveletTransformATrousAlgorithm')
else:
warnings.warn(
"The python package pySAP is not used for starlet operations. "
"They will be performed using (slower) python routines.")
self._fast_inverse = fast_inverse
self._second_gen = second_gen
self._show_pysap_plots = show_pysap_plots
self.interpol = Interpol()
self.thread_count = thread_count
def function(self,
x,
y,
amp=None,
n_scales=None,
n_pixels=None,
scale=1,
center_x=0,
center_y=0):
"""
1D inverse starlet transform from starlet coefficients stored in coeffs
Follows lenstronomy conventions for light profiles.
:param amp: decomposition coefficients ('amp' to follow conventions in other light profile)
This is an ndarray with shape (n_scales, sqrt(n_pixels), sqrt(n_pixels)) or (n_scales*n_pixels,)
:param n_scales: number of decomposition scales
:param n_pixels: number of pixels in a single scale
:return: reconstructed signal as 1D array of shape (n_pixels,)
"""
if len(amp.shape) == 1:
coeffs = util.array2cube(amp, n_scales, n_pixels)
elif len(amp.shape) == 3:
coeffs = amp
else:
raise ValueError(
"Starlets 'amp' has not the right shape (1D or 3D arrays are supported)"
)
image = self.function_2d(coeffs, n_scales, n_pixels)
image = self.interpol.function(x,
y,
image=image,
scale=scale,
center_x=center_x,
center_y=center_y,
amp=1,
phi_G=0)
return image
def function_2d(self, coeffs, n_scales, n_pixels):
"""
2D inverse starlet transform from starlet coefficients stored in coeffs
:param coeffs: decomposition coefficients,
ndarray with shape (n_scales, sqrt(n_pixels), sqrt(n_pixels))
:param n_scales: number of decomposition scales
:return: reconstructed signal as 2D array of shape (sqrt(n_pixels), sqrt(n_pixels))
"""
if self.use_pysap and not self._second_gen:
return self._inverse_transform(coeffs, n_scales, n_pixels)
else:
return starlets_util.inverse_transform(coeffs,
fast=self._fast_inverse,
second_gen=self._second_gen)
def decomposition(self, image, n_scales):
"""
1D starlet transform from starlet coefficients stored in coeffs
:param image: 2D image to be decomposed, ndarray with shape (sqrt(n_pixels), sqrt(n_pixels))
:param n_scales: number of decomposition scales
:return: reconstructed signal as 1D array of shape (n_scales*n_pixels,)
"""
if len(image.shape) == 1:
image_2d = util.array2image(image)
elif len(image.shape) == 2:
image_2d = image
else:
raise ValueError(
"image has not the right shape (1D or 2D arrays are supported for starlets decomposition)"
)
return util.cube2array(self.decomposition_2d(image_2d, n_scales))
def decomposition_2d(self, image, n_scales):
"""
2D starlet transform from starlet coefficients stored in coeffs
:param image: 2D image to be decomposed, ndarray with shape (sqrt(n_pixels), sqrt(n_pixels))
:param n_scales: number of decomposition scales
:return: reconstructed signal as 2D array of shape (n_scales, sqrt(n_pixels), sqrt(n_pixels))
"""
if self.use_pysap and not self._second_gen:
coeffs = self._transform(image, n_scales)
else:
coeffs = starlets_util.transform(image,
n_scales,
second_gen=self._second_gen)
return coeffs
def _inverse_transform(self, coeffs, n_scales, n_pixels):
"""reconstructs image from starlet coefficients"""
self._check_transform_pysap(n_scales, n_pixels)
if self._fast_inverse and not self._second_gen:
# for 1st gen starlet the reconstruction can be performed by summing all scales
image = np.sum(coeffs, axis=0)
else:
coeffs = self._coeffs2pysap(coeffs)
self._transf.analysis_data = coeffs
result = self._transf.synthesis()
if self._show_pysap_plots:
result.show()
image = result.data
return image
def _transform(self, image, n_scales):
"""decomposes an image into starlets coefficients"""
self._check_transform_pysap(n_scales, image.size)
self._transf.data = image
self._transf.analysis()
if self._show_pysap_plots:
self._transf.show()
coeffs = self._transf.analysis_data
coeffs = self._pysap2coeffs(coeffs)
return coeffs
def _check_transform_pysap(self, n_scales, n_pixels):
"""if needed, update the loaded pySAP transform to correct number of scales"""
if not hasattr(
self, '_transf'
) or n_scales != self._n_scales or n_pixels != self._n_pixels:
self._transf = self._transf_class(nb_scale=n_scales,
verbose=False,
nb_procs=self.thread_count)
self._n_scales = n_scales
self._n_pixels = n_pixels
def _pysap2coeffs(self, coeffs):
"""convert pySAP decomposition coefficients to numpy array"""
return np.asarray(coeffs)
def _coeffs2pysap(self, coeffs):
"""convert coefficients stored in numpy array to list required by pySAP"""
coeffs_list = []
for i in range(coeffs.shape[0]):
coeffs_list.append(coeffs[i, :, :])
return coeffs_list
def _load_pysap(self, force_no_pysap):
"""load pySAP module"""
if force_no_pysap:
return False, None
try:
import pysap
except ImportError:
return False, None
else:
return True, pysap
def delete_cache(self):
"""delete the cached interpolated image"""
self.interpol.delete_cache()