# coding: utf-8
from typing import Optional, Callable, Union, Tuple, Iterable
import numpy as np
from pathlib import Path
from .camera_processes import GPUCorrelProcess
from .camera import Camera
[docs]
class GPUCorrel(Camera):
"""This Block can perform GPU-accelerated image correlation on images
acquired by a :class:`~crappy.camera.Camera` object, and project the result
on various fields.
It is mostly used for computing the displacement and the strain over the
given patch, but other fields are also available. refer to the ``fields`` and
``labels`` arguments for more details.
This Block takes no input :class:`~crappy.links.Link` in a majority of
situations, and outputs the results of image correlation. It is a subclass of
the :class:`~crappy.blocks.Camera` Block, and inherits of all its features.
That includes the possibility to record and to display images in real-time,
simultaneously to the image acquisition and processing. Refer to the
documentation of the Camera Block for more information on these features.
This Block is very similar to the :class:`DISCorrel` Block, except this
latter uses DISFlow to perform the image correlation and is not
GPU-accelerated. The :class:`~crappy.blocks.GPUVE` Block also relies on
GPU-accelerated image correlation for computing the displacement and strain
on images, but it tracks multiple patches and uses video-extensometry.
No Region Of Interest can be specified to this Block, so by default the
correlation is performed on the entire image. It is however possible to set
a mask, so that only part of the image is considered when running the
correlation.
Warning:
This Block cannot run with CUDA versions greater than 11.3 ! This is due to
a deprecation in pycuda, and is unlikely to be fixed anytime soon in Crappy
or pycuda.
.. versionadded:: 1.4.0
"""
[docs]
def __init__(self,
camera: str,
fields: Union[str, Iterable[str]],
img_shape: Tuple[int, int],
img_dtype: str,
transform: Optional[Callable[[np.ndarray], np.ndarray]] = None,
display_images: bool = False,
displayer_backend: Optional[str] = None,
displayer_framerate: float = 5,
software_trig_label: Optional[str] = None,
verbose: int = 0,
freq: Optional[float] = 200,
debug: Optional[bool] = False,
save_images: bool = False,
img_extension: str = "tiff",
save_folder: Optional[Union[str, Path]] = None,
save_period: int = 1,
save_backend: Optional[str] = None,
image_generator: Optional[Callable[[float, float],
np.ndarray]] = None,
labels: Optional[Union[str, Iterable[str]]] = None,
discard_limit: float = 3,
discard_ref: int = 5,
img_ref: Optional[np.ndarray] = None,
levels: int = 5,
resampling_factor: float = 2,
kernel_file: Optional[Union[str, Path]] = None,
iterations: int = 4,
mask: Optional[np.ndarray] = None,
mul: float = 3,
res: bool = False,
**kwargs) -> None:
"""Sets the arguments and initializes the parent class.
Args:
camera: The name of the :class:`~crappy.camera.Camera` object to use for
acquiring the images. Arguments can be passed to this Camera as
``kwargs`` of this Block. This argument is ignored if the
``image_generator`` argument is provided.
fields: The several fields to calculate on the acquired images. They
should be given as an iterable containing :obj:`str`. Each string
represents one field to calculate, so the more fields are given the
heavier the computation is. The possible fields are :
::
'x', 'y', 'r', 'exx', 'eyy', 'exy', 'eyx', 'exy2', 'z'
For each field, one single value is computed, corresponding to the
norm of the field values over the patch area.
img_shape: The shape of the images returned by the
:class:`~crappy.camera.Camera` object as a :obj:`tuple` of :obj:`int`.
It should correspond to the value returned by :obj:`numpy.shape`.
img_dtype: The `dtype` of the images returned by the
:class:`~crappy.camera.Camera` object, as a :obj:`str`. It should
correspond to a valid data type in :mod:`numpy`, e.g. ``'uint8'``.
transform: A callable taking an image as an argument, and returning a
transformed image as an output. Allows applying a post-processing
operation to the acquired images. This is done right after the
acquisition, so the original image is permanently lost and only the
transformed image is displayed and/or saved and/or further processed.
The transform operation is not parallelized, so it might negatively
affect the acquisition framerate if it is too heavy.
.. versionadded:: 1.5.10
display_images: If :obj:`True`, displays the acquired images in a
dedicated window, using the backend given in ``displayer_backend`` and
at the frequency specified in ``displayer_framerate``. This option
should be considered as a debug or basic follow-up feature, it is not
intended to be very fast nor to display high-quality images. The
maximum resolution of the displayed images in `640x480`, the images
might be downscaled to fit in this format.
.. versionchanged:: 1.5.10
renamed from *show_image* to *display_images*
displayer_backend: The backend to use for displaying the images. Can be
either ``'cv2'`` or ``'mpl'``, to use respectively :mod:`cv2` (OpenCV)
or :mod:`matplotlib`. ``'cv2'`` usually allows achieving a higher
display frequency. Ignored if ``display_images`` is :obj:`False`. If
not given and ``display_images`` is :obj:`True`, ``'cv2'`` is tried
first and ``'mpl'`` second, and the first available one is used.
.. versionadded:: 1.5.10
displayer_framerate: The maximum update frequency of the image displayer,
as an :obj:`int`. This value usually lies between 5 and 30Hz, the
default is 5. The achieved update frequency might be lower than
requested. Ignored if ``display_images`` is :obj:`False`.
.. versionadded:: 1.5.10
software_trig_label: The name of a label used as a software trigger for
the :class:`~crappy.camera.Camera`. If given, images will only be
acquired when receiving data over this label. The received value does
not matter. This software trigger is not meant to be very precise, it
is recommended not to rely on it for a trigger frequency greater than
10Hz, in which case a hardware trigger should be preferred if available
on the camera.
.. versionadded:: 2.0.0
verbose: The verbose level as an integer, between `0` and `3`. At level
`0` no information is displayed, and at level `3` so much information
is displayed that it slows the code down. This argument allows to
adjust the precision of the log messages, while the ``debug`` argument
is for enabling or disabling logging.
freq: The target looping frequency for the Block. If :obj:`None`, loops
as fast as possible.
.. versionadded:: 1.5.10
debug: If :obj:`True`, displays all the log messages including the
:obj:`~logging.DEBUG` ones. If :obj:`False`, only displays the log
messages with :obj:`~logging.INFO` level or higher. If :obj:`None`,
disables logging for this Block.
.. versionadded:: 2.0.0
save_images: If :obj:`True`, the acquired images are saved to the folder
specified in ``save_folder``, in the format specified in
``img_extension``, using the backend specified in ``save_backend``, and
at the frequency specified in ``save_period``. Each image is saved with
the name : ``<frame_nr>_<timestamp>.<extension>``, and can thus easily
be identified. Along with the images, a ``metadata.csv`` file records
the metadata of all the saved images. This metadata is either the one
returned by the :meth:`~crappy.camera.Camera.get_image` method of the
:class:`~crappy.camera.Camera` object, or the default one generated in
the :meth:`~crappy.blocks.Camera.loop` method of the
:class:`~crappy.blocks.Camera` Block. Depending on the framerate of the
camera and the performance of the computer, it is not guaranteed that
all the acquired images will be recorded.
.. versionadded:: 1.5.10
img_extension: The file extension for the recorded images, as a
:obj:`str` and without the dot. Common file extensions include `tiff`,
`png`, `jpg`, etc. Depending on the used ``save_backend``, some
extensions might not be available. It is currently not possible to
customize the save parameters further than choosing the file extension.
Ignored if ``save_images`` is :obj:`False`.
.. versionadded:: 2.0.0
save_folder: Path to the folder where to save the images, either as a
:obj:`str` or as a :obj:`pathlib.Path`. Can be an absolute or a
relative path, pointing to a folder. If the folder does not exist, it
will be created (if the user has permission). If the given folder
already contains a ``metadata.csv`` file (and thus likely images from
Crappy), images are saved to another folder with the same name except
a suffix is appended. Ignored if ``save_images`` is :obj:`False`. If
not provided and ``save_images`` is :obj:`True`, the images are saved
to the folder ``Crappy_images``, created next to the running script.
.. versionadded:: 1.5.10
save_period: Must be given as an :obj:`int`. Only one out of that number
images at most will be saved. Allows to have a known periodicity in
case the framerate is too high to record all the images. Or simply to
reduce the number of recorded images if saving them all is not needed.
Ignored if ``save_images`` is :obj:`False`.
.. versionadded:: 1.5.10
save_backend: If ``save_images`` is :obj:`True`, the backend to use for
recording the images. It should be one of:
::
'sitk', 'cv2', 'pil', 'npy'
They correspond to the modules :mod:`SimpleITK`, :mod:`cv2` (OpenCV),
:mod:`PIL` (Pillow Fork), and :mod:`numpy`. Note that the ``'npy'``
backend saves the images as raw :obj:`numpy.array`, and thus ignores
the ``img_extension`` argument. Depending on the machine, some backends
may be faster or slower. For using each backend, the corresponding
Python must of course be installed. If not provided and ``save_images``
is :obj:`True`, the backends are tried in the same order as given above
and the first available one is used. ``'npy'`` is always available.
.. versionadded:: 1.5.10
image_generator: A callable taking two :obj:`float` as arguments and
returning an image as a :obj:`numpy.array`. **This argument is intended
for use in the examples of Crappy, to apply an artificial strain on a
base image. Most users should ignore it.** When given, the ``camera``
argument is ignored and the images are acquired from the generator. To
apply a strain on the image, strain values (in `%`) should be sent to
the Camera Block over the labels ``'Exx(%)'`` and ``'Eyy(%)'``.
.. versionadded:: 1.5.10
labels: The labels to use for sending data to downstream Blocks. If not
given, the default labels are ``'t(s)', 'meta'`` followed by the names
of the given ``fields``. They carry for each image its timestamp, a
:obj:`dict` containing its metadata, and then for each field the
computed value as a :obj:`float`. When setting this argument, make sure
to give at least 2 labels for the time and the metadata, and one label
per field. The ``'res'`` label containing the residuals if ``res`` is
:obj:`True` should not be included here, it will be automatically
added.
discard_limit: If ``res`` is :obj:`True`, the result of the
correlation is not sent to the downstream Blocks if the residuals for
the current image are greater than ``discard_limit`` times the average
residual for the last ``discard_ref`` images.
.. versionchanged:: 1.5.10
renamed from *discard_lim* to *discard_limit*
discard_ref: If ``res`` is :obj:`True`, the result of the
correlation is not sent to the downstream Blocks if the residuals for
the current image are greater than ``discard_limit`` times the average
residual for the last ``discard_ref`` images.
img_ref: A reference image, as a 2D :obj:`numpy.array` with `dtype`
`float32`. If given, it will be set early on the correlation class and
the test can start from the first acquired frame. If not given, the
first acquired image will be set as the reference, which will slow down
the beginning of the test.
.. versionchanged:: 1.5.10 renamed from *imgref* to *img_ref*
levels: Number of levels of the pyramid. More levels may help converging
on images with large strain, but may fail on images that don't contain
low spatial frequency. Fewer levels mean that the program runs faster.
.. versionadded:: 1.5.10
resampling_factor: The factor by which the resolution is divided between
each stage of the pyramid. A low factor ensures coherence between the
stages, but is more computationally intensive. A high factor allows
reaching a finer detail level, but may lead to a coherence loss between
the stages.
.. versionadded:: 1.5.10
kernel_file: The path to the file containing the kernels to use for the
correlation. Can be a :obj:`pathlib.Path` object or a :obj:`str`. If
not provided, the default :ref:`GPU Kernels` are used.
.. versionadded:: 1.5.10
iterations: The maximum number of iterations to run before returning the
results. The results may be returned before if the residuals start
increasing.
.. versionadded:: 1.5.10
mask: The mask used for weighting the region of interest on the image. It
is generally used to prevent unexpected behavior on the border of the
image. Also allows to select a sub-region of the image if the
correlation should not be performed on the entire image, as this Block
does not accept a `patch` argument.
.. versionadded:: 1.5.10
mul: The scalar by which the direction will be multiplied before being
added to the solution. If it's too high, the convergence will be fast
but there's a risk to go past the solution and to diverge. If it's too
low, the convergence will be slower and require more iterations. `3`
was found to be an acceptable value in most cases, but it is
recommended to tune this value for each application so that the
convergence is neither too slow nor too fast.
.. versionadded:: 1.5.10
res: If :obj:`True`, calculates the residuals after performing the
correlation and returns the residuals along with the correlation data.
The residuals are always returned under the label ``'res'``, and this
label should not be included in the ``labels`` argument.
.. versionadded:: 1.5.10
**kwargs: Any additional argument will be passed to the
:class:`~crappy.camera.Camera` object, and used as a kwarg to its
:meth:`~crappy.camera.Camera.open` method.
.. versionremoved:: 1.5.10
*fps_label*, *ext*, *input_label*, *config* and *cam_kwargs* arguments
.. versionremoved:: 2.0.0 *img_name* argument
"""
super().__init__(camera=camera,
transform=transform,
config=False,
display_images=display_images,
displayer_backend=displayer_backend,
displayer_framerate=displayer_framerate,
software_trig_label=software_trig_label,
display_freq=bool(verbose),
freq=freq,
debug=debug,
save_images=save_images,
img_extension=img_extension,
save_folder=save_folder,
save_period=save_period,
save_backend=save_backend,
image_generator=image_generator,
img_shape=img_shape,
img_dtype=img_dtype,
**kwargs)
# Forcing the fields into a list
if isinstance(fields, str):
fields = [fields]
else:
fields = list(fields)
# Forcing the labels into a list
if labels is None:
self.labels = ['t(s)', 'meta'] + fields
elif isinstance(labels, str):
self.labels = [labels]
else:
self.labels = list(labels)
if res:
self.labels.append('res')
self._calc_res = res
# Checking that the number of fields and the number of labels match
if 2 + len(fields) + int(res) != len(self.labels):
raise ValueError("The number of fields is inconsistent with the number "
"of labels !\nMake sure that the time label was given")
# These arguments are for the GPUCorrelProcess
self._discard_limit = discard_limit
self._discard_ref = discard_ref
self._img_ref = img_ref
self._verbose = verbose
self._levels = levels
self._resampling_factor = resampling_factor
self._kernel_file = kernel_file
self._iterations = iterations
self._fields = fields
self._mask = mask
self._mul = mul
[docs]
def prepare(self) -> None:
"""This method mostly calls the :meth:`~crappy.blocks.Camera.prepare`
method of the parent class.
In addition to that it instantiates the
:class:`~crappy.blocks.camera_processes.GPUCorrelProcess` object that
performs the GPU-accelerated image correlation.
.. versionchanged:: 1.5.5 now accepting args and kwargs
.. versionchanged:: 1.5.10 not accepting arguments anymore
"""
# Instantiating the GPUCorrelProcess
self.process_proc = GPUCorrelProcess(
discard_limit=self._discard_limit,
discard_ref=self._discard_ref,
calc_res=self._calc_res,
img_ref=self._img_ref,
verbose=self._verbose,
levels=self._levels,
resampling_factor=self._resampling_factor,
kernel_file=self._kernel_file,
iterations=self._iterations,
fields=self._fields,
mask=self._mask,
mul=self._mul)
super().prepare()
def _configure(self) -> None:
"""No configuration window is available for this Block, so this method was
left blank."""
...