Inheritance diagram for lsst.scarlet.lite.observation.Observation:
Public Member Functions | |
| __init__ (self, np.ndarray|Image images, np.ndarray|Image variance, np.ndarray|Image weights, np.ndarray psfs, np.ndarray|None model_psf=None, np.ndarray|None noise_rms=None, Box|None bbox=None, tuple|None bands=None, int padding=3, str convolution_mode="fft") | |
| tuple | bands (self) |
| Box | bbox (self) |
| Image | convolve (self, Image image, str|None mode=None, bool grad=False) |
| float | log_likelihood (self, Image model) |
| tuple[int, int, int] | shape (self) |
| int | n_bands (self) |
| npt.DTypeLike | dtype (self) |
| tuple[int, int, int, int] | convolution_bounds (self) |
Static Public Member Functions | |
| Observation | empty (tuple[Any] bands, np.ndarray psfs, np.ndarray model_psf, Box bbox, npt.DTypeLike dtype) |
Public Attributes | |
| images | |
| variance | |
| weights | |
| psfs | |
| mode | |
| noise_rms | |
| model_psf | |
| padding | |
| diff_kernel | |
| grad_kernel | |
Protected Attributes | |
| _convolution_bounds | |
Detailed Description
A single observation
This class contains all of the observed images and derived
properties, like PSFs, variance map, and weight maps,
required for most optimizers.
This includes methods to match a scarlet model PSF to the oberved PSF
in each band.
Notes
-----
This is effectively a combination of the `Observation` and
`Renderer` class from scarlet main, greatly simplified due
to the assumptions that the observations are all resampled
onto the same pixel grid and that the `images` contain all
of the information for all of the model bands.
Parameters
----------
images:
(bands, y, x) array of observed images.
variance:
(bands, y, x) array of variance for each image pixel.
weights:
(bands, y, x) array of weights to use when calculate the
likelihood of each pixel.
psfs:
(bands, y, x) array of the PSF image in each band.
model_psf:
(bands, y, x) array of the model PSF image in each band.
If `model_psf` is `None` then convolution is performed,
which should only be done when the observation is a
PSF matched coadd, and the scarlet model has the same PSF.
noise_rms:
Per-band average noise RMS. If `noise_rms` is `None` then the mean
of the sqrt of the variance is used.
bbox:
The bounding box containing the model. If `bbox` is `None` then
a `Box` is created that is the shape of `images` with an origin
at `(0, 0)`.
padding:
Padding to use when performing an FFT convolution.
convolution_mode:
The method of convolution. This should be either "fft" or "real".
Definition at line 166 of file observation.py.
Constructor & Destructor Documentation
◆ __init__()
| lsst.scarlet.lite.observation.Observation.__init__ | ( | self, | |
| np.ndarray | Image | images, | ||
| np.ndarray | Image | variance, | ||
| np.ndarray | Image | weights, | ||
| np.ndarray | psfs, | ||
| np.ndarray | None | model_psf = None, | ||
| np.ndarray | None | noise_rms = None, | ||
| Box | None | bbox = None, | ||
| tuple | None | bands = None, | ||
| int | padding = 3, | ||
| str | convolution_mode = "fft" ) |
Reimplemented in lsst.scarlet.lite.models.fit_psf.FittedPsfObservation.
Definition at line 212 of file observation.py.
224 ):
225 # Convert the images to a multi-band `Image` and use the resulting
226 # bbox and bands.
227 images = _set_image_like(images, bands, bbox)
228 bands = images.bands
229 bbox = images.bbox
230 self.images = images
231 self.variance = _set_image_like(variance, bands, bbox)
232 self.weights = _set_image_like(weights, bands, bbox)
233 # make sure that the images and psfs have the same dtype
234 if psfs.dtype != images.dtype:
235 psfs = psfs.astype(images.dtype)
236 self.psfs = psfs
237
238 if convolution_mode not in [
239 "fft",
240 "real",
241 ]:
242 raise ValueError("convolution_mode must be either 'fft' or 'real'")
243 self.mode = convolution_mode
244 if noise_rms is None:
245 noise_rms = np.array(np.mean(np.sqrt(variance.data), axis=(1, 2)))
246 self.noise_rms = noise_rms
247
248 # Create a difference kernel to convolve the model to the PSF
249 # in each band
250 self.model_psf = model_psf
251 self.padding = padding
252 if model_psf is not None:
253 if model_psf.dtype != images.dtype:
254 self.model_psf = model_psf.astype(images.dtype)
255 self.diff_kernel: Fourier | None = cast(Fourier, match_kernel(psfs, model_psf, padding=padding))
256 # The gradient of a convolution is another convolution,
257 # but with the flipped and transposed kernel.
258 diff_img = self.diff_kernel.image
259 self.grad_kernel: Fourier | None = Fourier(diff_img[:, ::-1, ::-1])
260 else:
261 self.diff_kernel = None
262 self.grad_kernel = None
263
264 self._convolution_bounds: tuple[int, int, int, int] | None = None
265
Member Function Documentation
◆ bands()
| tuple lsst.scarlet.lite.observation.Observation.bands | ( | self | ) |
The bands in the observations.
Definition at line 267 of file observation.py.
267 def bands(self) -> tuple:
268 """The bands in the observations."""
269 return self.images.bands
270
◆ bbox()
| Box lsst.scarlet.lite.observation.Observation.bbox | ( | self | ) |
The bounding box for the full observation.
Definition at line 272 of file observation.py.
273 """The bounding box for the full observation."""
274 return self.images.bbox
275
◆ convolution_bounds()
| tuple[int, int, int, int] lsst.scarlet.lite.observation.Observation.convolution_bounds | ( | self | ) |
Build the slices needed for convolution in real space
Definition at line 367 of file observation.py.
367 def convolution_bounds(self) -> tuple[int, int, int, int]:
368 """Build the slices needed for convolution in real space"""
369 if self._convolution_bounds is None:
370 coords = get_filter_coords(cast(Fourier, self.diff_kernel).image[0])
371 self._convolution_bounds = get_filter_bounds(coords.reshape(-1, 2))
372 return self._convolution_bounds
373
◆ convolve()
| Image lsst.scarlet.lite.observation.Observation.convolve | ( | self, | |
| Image | image, | ||
| str | None | mode = None, | ||
| bool | grad = False ) |
Convolve the model into the observed seeing in each band.
Parameters
----------
image:
The 3D image to convolve.
mode:
The convolution mode to use.
This should be "real" or "fft" or `None`,
where `None` will use the default `convolution_mode`
specified during init.
grad:
Whether this is a backward gradient convolution
(`grad==True`) or a pure convolution with the PSF.
Returns
-------
result:
The convolved image.
Reimplemented in lsst.scarlet.lite.models.fit_psf.FittedPsfObservation.
Definition at line 276 of file observation.py.
276 def convolve(self, image: Image, mode: str | None = None, grad: bool = False) -> Image:
277 """Convolve the model into the observed seeing in each band.
278
279 Parameters
280 ----------
281 image:
282 The 3D image to convolve.
283 mode:
284 The convolution mode to use.
285 This should be "real" or "fft" or `None`,
286 where `None` will use the default `convolution_mode`
287 specified during init.
288 grad:
289 Whether this is a backward gradient convolution
290 (`grad==True`) or a pure convolution with the PSF.
291
292 Returns
293 -------
294 result:
295 The convolved image.
296 """
297 if grad:
298 kernel = self.grad_kernel
299 else:
300 kernel = self.diff_kernel
301
302 if kernel is None:
303 return image
304
305 if mode is None:
306 mode = self.mode
307 if mode == "fft":
308 result = fft_convolve(
309 Fourier(image.data),
310 kernel,
311 axes=(1, 2),
312 return_fourier=False,
313 )
314 elif mode == "real":
315 dy = image.shape[1] - kernel.image.shape[1]
316 dx = image.shape[2] - kernel.image.shape[2]
317 if dy < 0 or dx < 0:
318 # The image needs to be padded because it is smaller than
319 # the psf kernel
320 _image = image.data
321 newshape = list(_image.shape)
322 if dy < 0:
323 newshape[1] += kernel.image.shape[1] - image.shape[1]
324 if dx < 0:
325 newshape[2] += kernel.image.shape[2] - image.shape[2]
326 _image = _pad(_image, newshape)
327 result = convolve(_image, kernel.image, self.convolution_bounds)
328 result = centered(result, image.data.shape) # type: ignore
329 else:
330 result = convolve(image.data, kernel.image, self.convolution_bounds)
331 else:
332 raise ValueError(f"mode must be either 'fft' or 'real', got {mode}")
333 return Image(cast(np.ndarray, result), bands=image.bands, yx0=image.yx0)
334
◆ dtype()
| npt.DTypeLike lsst.scarlet.lite.observation.Observation.dtype | ( | self | ) |
The dtype of the observation is the dtype of the images
Definition at line 362 of file observation.py.
362 def dtype(self) -> npt.DTypeLike:
363 """The dtype of the observation is the dtype of the images"""
364 return self.images.dtype
365
◆ empty()
|
static |
Definition at line 375 of file observation.py.
377 ) -> Observation:
378 dummy_image = np.zeros((len(bands),) + bbox.shape, dtype=dtype)
379
380 return Observation(
381 images=dummy_image,
382 variance=dummy_image,
383 weights=dummy_image,
384 psfs=psfs,
385 model_psf=model_psf,
386 noise_rms=np.zeros((len(bands),), dtype=dtype),
387 bbox=bbox,
388 bands=bands,
389 convolution_mode="real",
390 )
◆ log_likelihood()
| float lsst.scarlet.lite.observation.Observation.log_likelihood | ( | self, | |
| Image | model ) |
Calculate the log likelihood of the given model
Parameters
----------
model:
Model to compare with the observed images.
Returns
-------
result:
The log-likelihood of the given model.
Definition at line 335 of file observation.py.
335 def log_likelihood(self, model: Image) -> float:
336 """Calculate the log likelihood of the given model
337
338 Parameters
339 ----------
340 model:
341 Model to compare with the observed images.
342
343 Returns
344 -------
345 result:
346 The log-likelihood of the given model.
347 """
348 result = 0.5 * -np.sum((self.weights * (self.images - model) ** 2).data)
349 return result
350
◆ n_bands()
| int lsst.scarlet.lite.observation.Observation.n_bands | ( | self | ) |
The number of bands in the observation
Definition at line 357 of file observation.py.
357 def n_bands(self) -> int:
358 """The number of bands in the observation"""
359 return self.images.shape[0]
360
◆ shape()
| tuple[int, int, int] lsst.scarlet.lite.observation.Observation.shape | ( | self | ) |
The shape of the images, variance, etc.
Definition at line 352 of file observation.py.
352 def shape(self) -> tuple[int, int, int]:
353 """The shape of the images, variance, etc."""
354 return cast(tuple[int, int, int], self.images.shape)
355
Member Data Documentation
◆ _convolution_bounds
|
protected |
Definition at line 371 of file observation.py.
◆ diff_kernel
| lsst.scarlet.lite.observation.Observation.diff_kernel |
Definition at line 261 of file observation.py.
◆ grad_kernel
| lsst.scarlet.lite.observation.Observation.grad_kernel |
Definition at line 262 of file observation.py.
◆ images
| lsst.scarlet.lite.observation.Observation.images |
Definition at line 230 of file observation.py.
◆ mode
| lsst.scarlet.lite.observation.Observation.mode |
Definition at line 243 of file observation.py.
◆ model_psf
| lsst.scarlet.lite.observation.Observation.model_psf |
Definition at line 250 of file observation.py.
◆ noise_rms
| lsst.scarlet.lite.observation.Observation.noise_rms |
Definition at line 246 of file observation.py.
◆ padding
| lsst.scarlet.lite.observation.Observation.padding |
Definition at line 251 of file observation.py.
◆ psfs
| lsst.scarlet.lite.observation.Observation.psfs |
Definition at line 236 of file observation.py.
◆ variance
| lsst.scarlet.lite.observation.Observation.variance |
Definition at line 231 of file observation.py.
◆ weights
| lsst.scarlet.lite.observation.Observation.weights |
Definition at line 232 of file observation.py.
The documentation for this class was generated from the following file:
- /j/snowflake/release/lsstsw/stack/lsst-scipipe-9.0.0/Linux64/scarlet_lite/g168dd56ebc+a8ce1bb630/python/lsst/scarlet/lite/observation.py
