1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
# SPDX-License-Identifier: LGPL-3.0-only
import radler as rd
import numpy as np
# Tests for the VectorUniquePtrImageAccessor classe. This class is created in
# C++ and can't be created directly by Python. They are available in a
# WorkTableEntry so the tests use that class as a provider for the tested
# classes.
def test_vector_unique_ptr_image_accessor_append():
data = np.zeros((1, 1), dtype=np.float32)
entry = rd.WorkTableEntry()
vector = entry.psfs
assert len(vector) == 0
vector.append(data)
assert len(vector) == 1
def get_psf_rectangular(size, width, height):
"""Define a PSF with a rectangular shape"""
psf = np.zeros((size, size), dtype=np.float32)
psf[
size // 2 - height : size // 2 + height + 1,
size // 2 - width : size // 2 + width + 1,
] = 0.4
psf[size // 2, size // 2] = 1
return psf
def get_psf_cross(size, width, height):
"""Define a PSF with a cross shape"""
psf = np.zeros((size, size), dtype=np.float32)
psf[size // 2 - height : size // 2 + height + 1, size // 2] = 0.4
psf[size // 2, size // 2 - width : size // 2 + width + 1] = 0.4
psf[size // 2, size // 2] = 1
return psf
def get_subimage(center_point_x, center_point_y, interval, img):
"""Returns a square from the input image img, with center in (center_point_x, center_point_y) and width/height = 2 * interval"""
return img[
center_point_x - interval : center_point_x + interval + 1,
center_point_y - interval : center_point_y + interval + 1,
]
def check_values(current_psf, residual, psf_center, source_coords):
"""Asserts that the residual and PSF have the same shape"""
# Since the point source does not lie in the center of the region (a random shift is applied), the PSF and the residual need to be aligned to have a shape match
# Select the region in the residual image (one of the 9 in the grid) corresponding to the input PSF coordinates
subimage_residual = get_subimage(
psf_center[0], psf_center[1], 15, residual
)
# Select a 10x10 region centered around the point source
subimage_residual_centered = get_subimage(
source_coords[0] % 30, source_coords[1] % 30, 5, subimage_residual
)
# Only consider a 10x10 region in the center of the PSF
subimage_psf = get_subimage(
current_psf.shape[0] // 2, current_psf.shape[1] // 2, 5, current_psf
)
# If the shapes match, the sum between residual and PSF contains only the point source in its center
combined = subimage_residual_centered + subimage_psf
np.testing.assert_allclose(
combined[combined.shape[0] // 2, combined.shape[0] // 2],
1,
rtol=1e-5,
atol=1e-6,
)
# After its value is checked, we can reset the center point to 0 to make the next check easier
combined[5, 5] = 0
np.testing.assert_allclose(
combined, np.zeros_like(combined), rtol=1e-6, atol=1e-6
)
def test_direction_dependent_psfs():
"""
This test checks that each point is deconvolved with the closest PSF:
1. Place 9 PSF on a regular 3x3 grid.
2. Place 9 point sources on a semi-regular grid.
3. Run one iteration of deconvolution with multiple PSFs and parallel deconvolution.
4. Check in the residual that the correct PSF has been subtracted.
This test is also documented at https://confluence.skatelescope.org/display/SE/Test+DD+PSFs
"""
image_size = 90
psf_size = image_size // 3
# Define a grid of 9 PSFs
# The coordinates given in the work table are the center of each cell in the grid.
center_offset = psf_size // 2
psf_centers = np.zeros((9, 2), dtype=np.int64)
for i in range(3):
for k in range(3):
coord_x = i * image_size // 3 + center_offset
coord_y = k * image_size // 3 + center_offset
psf_centers[3 * i + k, :] = [coord_x, coord_y]
work_table = rd.WorkTable(psf_centers, 0, 0)
# Define 9 PSFs with different shapes
# psf_size = image_size # TODO: remove after MR !87 is merged
w1 = 2
w2 = 4
w3 = 8
entry = rd.WorkTableEntry()
direction_dependent_psfs = []
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, w2, w2)
) # Square
direction_dependent_psfs.append(
get_psf_cross(psf_size, w2, w2)
) # Symmetrical cross
direction_dependent_psfs.append(
get_psf_cross(psf_size, w1, w3)
) # Horizontal cross
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, w2, 0)
) # Vertical line
direction_dependent_psfs.append(
get_psf_cross(psf_size, w3, w1)
) # Vertical cross
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, w2, w1)
) # Vertical rectangle
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, 0, w2)
) # Horizontal line
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, 0, 0)
) # Point
direction_dependent_psfs.append(
get_psf_rectangular(psf_size, w1, w2)
) # Horizontal rectangle
# Define dirty image. There are 9 points, one in each of the 9 regions defined by the PSF grid.
# The point is not in the center of the grid, but a random shift is applied (the point will remain in its region)
residual = np.zeros((image_size, image_size), dtype=np.float32)
np.random.seed(10)
rand_interval = 4
source_coords = np.zeros((9, 2), dtype=np.int64)
for i in range(3):
for k in range(3):
point_offset = np.random.randint(-rand_interval, rand_interval)
coord_x = psf_centers[3 * i + k, 0] + point_offset
coord_y = psf_centers[3 * i + k, 1] + point_offset
source_coords[3 * i + k, :] = [coord_x, coord_y]
for p in source_coords:
residual[p[0], p[1]] = 1
# Initialize an empty model image
model = np.zeros((image_size, image_size), np.float32)
# Define work table entry
for psf in direction_dependent_psfs:
entry.psfs.append(psf)
entry.residual = residual
entry.model = model
entry.polarization = rd.Polarization.stokes_i
entry.image_weight = 1.0
work_table.add_entry(entry)
# Define deconvolution settings
settings = rd.Settings()
settings.algorithm_type = rd.AlgorithmType.generic_clean
settings.trimmed_image_width = image_size
settings.trimmed_image_height = image_size
settings.pixel_scale.x = 1.0
settings.pixel_scale.y = 1.0
settings.minor_iteration_count = 1
settings.minor_loop_gain = 1.0
# The settings for parallel deconvolution give the same number of subimages as the PSF grid
settings.parallel.grid_width = 3
settings.parallel.grid_height = 3
settings.parallel.max_threads = 1
# Run 1 iteration of deconvolution
radler_object = rd.Radler(settings, work_table, 0)
radler_object.perform(0)
# Verify that the correct PSF is applied in the corresponding region
check_values(
direction_dependent_psfs[0], residual, psf_centers[0], source_coords[0]
)
check_values(
direction_dependent_psfs[3], residual, psf_centers[1], source_coords[1]
)
check_values(
direction_dependent_psfs[6], residual, psf_centers[2], source_coords[2]
)
check_values(
direction_dependent_psfs[1], residual, psf_centers[3], source_coords[3]
)
check_values(
direction_dependent_psfs[4], residual, psf_centers[4], source_coords[4]
)
check_values(
direction_dependent_psfs[7], residual, psf_centers[5], source_coords[5]
)
check_values(
direction_dependent_psfs[2], residual, psf_centers[6], source_coords[6]
)
check_values(
direction_dependent_psfs[5], residual, psf_centers[7], source_coords[7]
)
check_values(
direction_dependent_psfs[8], residual, psf_centers[8], source_coords[8]
)
|
