Coverage for python/lsst/images/convolution_kernels.py: 51%
130 statements
« prev ^ index » next coverage.py v7.14.3, created at 2026-07-01 02:06 -0700
1# This file is part of lsst-images.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (https://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# Use of this source code is governed by a 3-clause BSD-style
10# license that can be found in the LICENSE file.
11from __future__ import annotations
13__all__ = (
14 "ConvolutionKernel",
15 "ConvolutionKernelSerializationModel",
16 "ImageBasisConvolutionKernel",
17 "ImageBasisConvolutionKernelSerializationModel",
18)
20from abc import ABC, abstractmethod
21from collections.abc import Iterable, Iterator, Sequence
22from typing import TYPE_CHECKING, Any, ClassVar, Literal
24import numpy as np
25import pydantic
27from ._geom import YX, Bounds, Box
28from ._image import Image
29from .fields import ChebyshevField, Field, FieldSerializationModel
30from .serialization import (
31 ArchiveTree,
32 ArrayReferenceModel,
33 InlineArrayModel,
34 InputArchive,
35 InvalidParameterError,
36 OutputArchive,
37)
39if TYPE_CHECKING:
40 try:
41 from lsst.afw.math import LinearCombinationKernel as LegacyLinearCombinationKernel
42 except ImportError:
43 type LegacyLinearCombinationKernel = Any # type: ignore[no-redef]
46# This may become a union in the future.
47type ConvolutionKernelSerializationModel = ImageBasisConvolutionKernelSerializationModel
50class ConvolutionKernel(ABC):
51 """An abstract base class for spatially-varying convolution kernels."""
53 @property
54 @abstractmethod
55 def bounds(self) -> Bounds:
56 """The region where this convolution kernel is valid
57 (`~lsst.images.Bounds`).
58 """
59 raise NotImplementedError()
61 @property
62 @abstractmethod
63 def kernel_bbox(self) -> Box:
64 """Bounding box of all images returned by `compute_kernel_image`
65 (`~lsst.images.Box`).
66 """
67 raise NotImplementedError()
69 @abstractmethod
70 def compute_kernel_image(self, *, x: int, y: int) -> Image:
71 """Evaluate the kernel at a point.
73 Parameters
74 ----------
75 x
76 Column position coordinate to evaluate at.
77 y
78 Row position coordinate to evaluate at.
80 Returns
81 -------
82 Image
83 An image of the kernel, centered on the center of the center pixel,
84 which is defined to be ``(0, 0)`` by the image's origin.
85 """
86 raise NotImplementedError()
88 @abstractmethod
89 def serialize(self, archive: OutputArchive[Any]) -> ConvolutionKernelSerializationModel:
90 """Serialize the kernel to an output archive.
92 Parameters
93 ----------
94 archive
95 Archive to write to.
96 """
97 raise NotImplementedError()
100class ImageBasisConvolutionKernel(ConvolutionKernel):
101 """A convolution kernel formed by a linear combination of images
102 multiplied by `~lsst.images.fields.BaseField` instances.
104 Parameters
105 ----------
106 basis
107 A 3-d array holding the kernel images each basis function, with shape
108 ``(n, height, width)``.
109 spatial
110 Iterable of `.fields.BaseField` of length ``basis.shape[0]``, holding
111 the spatial variation of each basis kernel.
112 center_y
113 Center of the basis kernels in the x dimension. Defaults to
114 ``height//2``.
115 center_x
116 Center of the basis kernels in the x dimension. Defaults to
117 ``width//2``.
118 """
120 def __init__(
121 self,
122 basis: np.ndarray,
123 spatial: Iterable[Field],
124 center_y: int | None = None,
125 center_x: int | None = None,
126 ):
127 self._spatial = tuple(spatial)
128 bounds: Bounds | None = None
129 for field in self._spatial:
130 if field.unit is not None: 130 ↛ 131line 130 didn't jump to line 131 because the condition on line 130 was never true
131 raise ValueError("Kernel spatial fields should not have units.")
132 if bounds is None:
133 bounds = field.bounds
134 else:
135 bounds = bounds.intersection(field.bounds)
136 if bounds is None: 136 ↛ 137line 136 didn't jump to line 137 because the condition on line 136 was never true
137 raise ValueError("Must have at least one basis function.")
138 self._bounds = bounds
139 self._basis = basis
140 if self._basis.ndim != 3: 140 ↛ 141line 140 didn't jump to line 141 because the condition on line 140 was never true
141 raise ValueError(f"Basis array must be 3-d; shape={self._basis.shape}.")
142 if len(self._spatial) != self._basis.shape[0]: 142 ↛ 143line 142 didn't jump to line 143 because the condition on line 142 was never true
143 raise ValueError(
144 f"Number of spatial fields ({len(self._spatial)}) "
145 f"does not match basis array shape ({self._basis.shape})."
146 )
147 if center_y is None: 147 ↛ 148line 147 didn't jump to line 148 because the condition on line 147 was never true
148 center_y = self._basis.shape[1] // 2
149 if center_x is None: 149 ↛ 150line 149 didn't jump to line 150 because the condition on line 149 was never true
150 center_x = self._basis.shape[2] // 2
151 self._kernel_bbox = Box.from_shape(self._basis.shape[1:], start=YX(y=-center_y, x=-center_x))
153 @property
154 def bounds(self) -> Bounds:
155 return self._bounds
157 @property
158 def kernel_bbox(self) -> Box:
159 return self._kernel_bbox
161 @property
162 def spatial(self) -> Sequence[Field]:
163 """The spatial variation of each basis function
164 (`~collections.abc.Sequence` [`~.fields.BaseField`]).
165 """
166 return self._spatial
168 @property
169 def basis(self) -> np.ndarray:
170 """The kernel basis functions, as an array with shape ``(n, h, w)``
171 (`numpy.ndarray`).
172 """
173 return self._basis
175 def __len__(self) -> int:
176 return len(self._spatial)
178 def __iter__(self) -> Iterator[tuple[Image, Field]]:
179 for field, array in zip(self._spatial, self._basis, strict=True):
180 yield Image(array, bbox=self._kernel_bbox), field
182 def compute_kernel_image(self, *, x: int, y: int) -> Image:
183 # TODO[DM-54965]: simplify this once BaseField.__call__ behaves more
184 # like a real ufunc and can handle scalars directly.
185 x_array = np.array([x], dtype=np.float64)
186 y_array = np.array([y], dtype=np.float64)
187 weights = np.array(
188 [spatial_field(x=x_array, y=y_array)[0] for spatial_field in self._spatial],
189 dtype=np.float64,
190 )
191 return Image(np.tensordot(weights, self._basis, axes=(0, 0)), bbox=self._kernel_bbox)
193 def serialize(self, archive: OutputArchive[Any]) -> ImageBasisConvolutionKernelSerializationModel:
194 """Serialize the kernel to an output archive.
196 Parameters
197 ----------
198 archive
199 Archive to write to.
200 """
201 serialized_basis = archive.add_array(self._basis, name="basis")
202 serialized_spatial = [archive.serialize_direct("spatial", f.serialize) for f in self._spatial]
203 return ImageBasisConvolutionKernelSerializationModel(
204 basis=serialized_basis,
205 spatial=serialized_spatial,
206 center_y=-self._kernel_bbox.y.min,
207 center_x=-self._kernel_bbox.x.min,
208 )
210 @staticmethod
211 def _get_archive_tree_type(
212 pointer_type: type[Any],
213 ) -> type[ImageBasisConvolutionKernelSerializationModel]:
214 """Return the serialization model type for this object for an archive
215 type that uses the given pointer type.
216 """
217 return ImageBasisConvolutionKernelSerializationModel
219 @staticmethod
220 def from_legacy(legacy_kernel: LegacyLinearCombinationKernel) -> ImageBasisConvolutionKernel:
221 """Convert from a legacy `lsst.afw.math.LinearCombinationKernel`.
223 Parameters
224 ----------
225 legacy_kernel
226 The kernel to convert. Must use Chebyshev polynomials for its
227 spatial variation and `lsst.afw.math.FixedKernel` objects with a
228 consistent shape and center for its basis functions.
229 """
230 from lsst.afw.math import FixedKernel as LegacyFixedKernel
231 from lsst.afw.math import LinearCombinationKernel as LegacyLinearCombinationKernel
233 if not isinstance(legacy_kernel, LegacyLinearCombinationKernel):
234 raise TypeError(
235 f"Cannot convert {type(legacy_kernel).__name__} instance to an ImageBasisConvolutionKernel."
236 )
237 dimensions = legacy_kernel.getDimensions()
238 center = legacy_kernel.getCtr()
239 basis = np.zeros((legacy_kernel.getNBasisKernels(), dimensions.y, dimensions.x), dtype=np.float64)
240 for n, basis_kernel in enumerate(legacy_kernel.getKernelList()):
241 if basis_kernel.getDimensions() != dimensions:
242 raise ValueError("Cannot convert LinearCombinationKernel with different-size basis kernels.")
243 if basis_kernel.getCtr() != center:
244 raise ValueError(
245 "Cannot convert LinearCombinationKernel with differently-centered basis kernels."
246 )
247 if not isinstance(basis_kernel, LegacyFixedKernel):
248 raise ValueError("Cannot convert LinearCombinationKernel with non-fixed basis kernels.")
249 legacy_image_view = Image(basis[n, :, :], dtype=np.float64).to_legacy()
250 basis_kernel.computeImage(legacy_image_view, doNormalize=False)
251 spatial = [ChebyshevField.from_legacy_function2(f) for f in legacy_kernel.getSpatialFunctionList()]
252 return ImageBasisConvolutionKernel(basis=basis, spatial=spatial, center_y=center.y, center_x=center.x)
254 def to_legacy(self) -> LegacyLinearCombinationKernel:
255 """Convert to a legacy `lsst.afw.math.LinearCombinationKernel`.
257 This only works if all spatial variation is handled by
258 `lsst.images.ChebyshevField`.
259 """
260 from lsst.afw.math import FixedKernel as LegacyFixedKernel
261 from lsst.afw.math import LinearCombinationKernel as LegacyLinearCombinationKernel
262 from lsst.geom import Point2I as LegacyPoint2I
264 basis_kernels = []
265 spatial_functions = []
266 legacy_center = LegacyPoint2I(-self._kernel_bbox.x.min, -self._kernel_bbox.y.min)
267 for image, field in self:
268 legacy_image = image.to_legacy()
269 legacy_image.setXY0(LegacyPoint2I())
270 basis_kernel = LegacyFixedKernel(legacy_image)
271 basis_kernel.setCtr(legacy_center)
272 basis_kernels.append(basis_kernel)
273 if not isinstance(field, ChebyshevField):
274 raise ValueError("Only Chebyshev spatial variation can be converted.")
275 spatial_functions.append(field.to_legacy_function2())
276 result = LegacyLinearCombinationKernel(basis_kernels, spatial_functions)
277 result.setCtr(legacy_center)
278 return result
281class ImageBasisConvolutionKernelSerializationModel(ArchiveTree):
282 """The serialization model for `ImageBasisConvolutionKernel`."""
284 SCHEMA_NAME: ClassVar[str] = "image_basis_convolution_kernel"
285 SCHEMA_VERSION: ClassVar[str] = "1.0.0"
286 MIN_READ_VERSION: ClassVar[int] = 1
287 PUBLIC_TYPE: ClassVar[type] = ImageBasisConvolutionKernel
289 basis: ArrayReferenceModel | InlineArrayModel = pydantic.Field(
290 description="The basis images, with shape (n, h, w)."
291 )
292 spatial: list[FieldSerializationModel] = pydantic.Field(
293 description="The spatial variation of each basis function."
294 )
295 center_y: int = pydantic.Field(description="Center row of the kernel in the basis images.")
296 center_x: int = pydantic.Field(description="Center column of the kernel in the basis images.")
298 kernel_type: Literal["IMAGE_BASIS"] = "IMAGE_BASIS"
300 def deserialize(self, archive: InputArchive[Any], **kwargs: Any) -> ImageBasisConvolutionKernel:
301 if kwargs: 301 ↛ 302line 301 didn't jump to line 302 because the condition on line 301 was never true
302 raise InvalidParameterError(f"Unrecognized parameters for ChebyshevField: {set(kwargs.keys())}.")
303 basis = archive.get_array(self.basis)
304 spatial = [f.deserialize(archive) for f in self.spatial]
305 return ImageBasisConvolutionKernel(basis, spatial, center_y=self.center_y, center_x=self.center_x)