kwarray.util_slider
¶
Module Contents¶
Classes¶
Slide a window of a certain shape over an array with a larger shape. |
|
Stitches multiple possibly overlapping slices into a larger array. |
Functions¶
|
Helper to generates slices in a single dimension. |
Attributes¶
- kwarray.util_slider.torch¶
- class kwarray.util_slider.SlidingWindow(shape, window, overlap=None, stride=None, keepbound=False, allow_overshoot=False)¶
Bases:
ubelt.NiceRepr
Slide a window of a certain shape over an array with a larger shape.
This can be used for iterating over a grid of sub-regions of 2d-images, 3d-volumes, or any n-dimensional array.
Yields slices of shape window that can be used to index into an array with shape shape via numpy / torch fancy indexing. This allows for fast fast iteration over subregions of a larger image. Because we generate a grid-basis using only shapes, the larger image does not need to be in memory as long as its width/height/depth/etc…
- Parameters
shape (Tuple[int, …]) – shape of source array to slide across.
window (Tuple[int, …]) – shape of window that will be slid over the larger image.
overlap (float, default=0) – a number between 0 and 1 indicating the fraction of overlap that parts will have. Specifying this is mutually exclusive with stride. Must be 0 <= overlap < 1.
stride (int, default=None) – the number of cells (pixels) moved on each step of the window. Mutually exclusive with overlap.
keepbound (bool, default=False) – if True, a non-uniform stride will be taken to ensure that the right / bottom of the image is returned as a slice if needed. Such a slice will not obey the overlap constraints. (Defaults to False)
allow_overshoot (bool, default=False) – if False, we will raise an error if the window doesn’t slide perfectly over the input shape.
- Variables
strides (basis_shape - shape of the grid corresponding to the number of) – the sliding window will take.
dimension (basis_slices - slices that will be taken in every) –
- Yields
Tuple[slice, …] –
- slices used for numpy indexing, the number of slices
in the tuple
Notes
For each dimension, we generate a basis (which defines a grid), and we slide over that basis.
Example
>>> from kwarray.util_slider import * # NOQA >>> shape = (10, 10) >>> window = (5, 5) >>> self = SlidingWindow(shape, window) >>> for i, index in enumerate(self): >>> print('i={}, index={}'.format(i, index)) i=0, index=(slice(0, 5, None), slice(0, 5, None)) i=1, index=(slice(0, 5, None), slice(5, 10, None)) i=2, index=(slice(5, 10, None), slice(0, 5, None)) i=3, index=(slice(5, 10, None), slice(5, 10, None))
Example
>>> from kwarray.util_slider import * # NOQA >>> shape = (16, 16) >>> window = (4, 4) >>> self = SlidingWindow(shape, window, overlap=(.5, .25)) >>> print('self.stride = {!r}'.format(self.stride)) self.stride = [2, 3] >>> list(ub.chunks(self.grid, 5)) [[(0, 0), (0, 1), (0, 2), (0, 3), (0, 4)], [(1, 0), (1, 1), (1, 2), (1, 3), (1, 4)], [(2, 0), (2, 1), (2, 2), (2, 3), (2, 4)], [(3, 0), (3, 1), (3, 2), (3, 3), (3, 4)], [(4, 0), (4, 1), (4, 2), (4, 3), (4, 4)], [(5, 0), (5, 1), (5, 2), (5, 3), (5, 4)], [(6, 0), (6, 1), (6, 2), (6, 3), (6, 4)]]
Example
>>> # Test shapes that dont fit >>> # When the window is bigger than the shape, the left-aligned slices >>> # are returend. >>> self = SlidingWindow((3, 3), (12, 12), allow_overshoot=True, keepbound=True) >>> print(list(self)) [(slice(0, 12, None), slice(0, 12, None))] >>> print(list(SlidingWindow((3, 3), None, allow_overshoot=True, keepbound=True))) [(slice(0, 3, None), slice(0, 3, None))] >>> print(list(SlidingWindow((3, 3), (None, 2), allow_overshoot=True, keepbound=True))) [(slice(0, 3, None), slice(0, 2, None)), (slice(0, 3, None), slice(1, 3, None))]
- __nice__(self)¶
- _compute_stride(self, overlap, stride, shape, window)¶
Ensures that stride hasoverlap the correct shape. If stride is not provided, compute stride from desired overlap.
- __len__(self)¶
- _iter_basis_frac(self)¶
- __iter__(self)¶
- __getitem__(self, index)¶
Get a specific item by its flat (raveled) index
Example
>>> from kwarray.util_slider import * # NOQA >>> window = (10, 10) >>> shape = (20, 20) >>> self = SlidingWindow(shape, window, stride=5) >>> itered_items = list(self) >>> assert len(itered_items) == len(self) >>> indexed_items = [self[i] for i in range(len(self))] >>> assert itered_items[0] == self[0] >>> assert itered_items[-1] == self[-1] >>> assert itered_items == indexed_items
- property grid(self)¶
Generate indices into the “basis” slice for each dimension. This enumerates the nd indices of the grid.
- Yields
Tuple[int, …]
- property slices(self)¶
Generate slices for each window (equivalent to iter(self))
Example
>>> shape = (220, 220) >>> window = (10, 10) >>> self = SlidingWindow(shape, window, stride=5) >>> list(self)[41:45] [(slice(0, 10, None), slice(205, 215, None)), (slice(0, 10, None), slice(210, 220, None)), (slice(5, 15, None), slice(0, 10, None)), (slice(5, 15, None), slice(5, 15, None))] >>> print('self.overlap = {!r}'.format(self.overlap)) self.overlap = [0.5, 0.5]
- property centers(self)¶
Generate centers of each window
- Yields
Tuple[float, …] – the center coordinate of the slice
Example
>>> shape = (4, 4) >>> window = (3, 3) >>> self = SlidingWindow(shape, window, stride=1) >>> list(zip(self.centers, self.slices)) [((1.0, 1.0), (slice(0, 3, None), slice(0, 3, None))), ((1.0, 2.0), (slice(0, 3, None), slice(1, 4, None))), ((2.0, 1.0), (slice(1, 4, None), slice(0, 3, None))), ((2.0, 2.0), (slice(1, 4, None), slice(1, 4, None)))] >>> shape = (3, 3) >>> window = (2, 2) >>> self = SlidingWindow(shape, window, stride=1) >>> list(zip(self.centers, self.slices)) [((0.5, 0.5), (slice(0, 2, None), slice(0, 2, None))), ((0.5, 1.5), (slice(0, 2, None), slice(1, 3, None))), ((1.5, 0.5), (slice(1, 3, None), slice(0, 2, None))), ((1.5, 1.5), (slice(1, 3, None), slice(1, 3, None)))]
- class kwarray.util_slider.Stitcher(stitcher, shape, device='numpy')¶
Bases:
ubelt.NiceRepr
Stitches multiple possibly overlapping slices into a larger array.
This is used to invert the SlidingWindow. For semenatic segmentation the patches are probability chips. Overlapping chips are averaged together.
- Parameters
shape (tuple) – dimensions of the large image that will be created from the smaller pixels or patches.
Todo
- [ ] Look at the old “add_fast” code in the netharn version and see if
it is worth porting. This code is kept in the dev folder in ../dev/_dev_slider.py
Example
>>> from kwarray.util_slider import * # NOQA >>> import sys >>> # Build a high resolution image and slice it into chips >>> highres = np.random.rand(5, 200, 200).astype(np.float32) >>> target_shape = (1, 50, 50) >>> slider = SlidingWindow(highres.shape, target_shape, overlap=(0, .5, .5)) >>> # Show how Sticher can be used to reconstruct the original image >>> stitcher = Stitcher(slider.input_shape) >>> for sl in list(slider): ... chip = highres[sl] ... stitcher.add(sl, chip) >>> assert stitcher.weights.max() == 4, 'some parts should be processed 4 times' >>> recon = stitcher.finalize()
- __nice__(stitcher)¶
- add(stitcher, indices, patch, weight=None)¶
Incorporate a new (possibly overlapping) patch or pixel using a weighted sum.
- Parameters
indices (slice or tuple) – typically a Tuple[slice] of pixels or a single pixel, but this can be any numpy fancy index.
patch (ndarray) – data to patch into the bigger image.
weight (float or ndarray) – weight of this patch (default to 1.0)
- average(stitcher)¶
Averages out contributions from overlapping adds using weighted average
- Returns
ndarray: the stitched image
- Return type
out
- finalize(stitcher, indices=None)¶
Averages out contributions from overlapping adds
- Parameters
indices (None | slice | tuple) – if None, finalize the entire block, otherwise only finalize a subregion.
- Returns
ndarray: the stitched image
- Return type
final
- kwarray.util_slider._slices1d(margin, stop, step=None, start=0, keepbound=False, check=True)¶
Helper to generates slices in a single dimension.
- Parameters
margin (int) – the length of the slice (window)
stop (int) – the length of the image dimension
step (int, default=None) – the length of each step / distance between slices
start (int, default=0) – starting point (in most cases set this to 0)
keepbound (bool) – if True, a non-uniform step will be taken to ensure that the right / bottom of the image is returned as a slice if needed. Such a slice will not obey the overlap constraints. (Defaults to False)
check (bool) – if True an error will be raised if the window does not cover the entire extent from start to stop, even if keepbound is True.
- Yields
slice – slice in one dimension of size (margin)
Example
>>> stop, margin, step = 2000, 360, 360 >>> keepbound = True >>> strides = list(_slices1d(margin, stop, step, keepbound, check=False)) >>> assert all([(s.stop - s.start) == margin for s in strides])
Example
>>> stop, margin, step = 200, 46, 7 >>> keepbound = True >>> strides = list(_slices1d(margin, stop, step, keepbound=False, check=True)) >>> starts = np.array([s.start for s in strides]) >>> stops = np.array([s.stop for s in strides]) >>> widths = stops - starts >>> assert np.all(np.diff(starts) == step) >>> assert np.all(widths == margin)
Example
>>> import pytest >>> stop, margin, step = 200, 36, 7 >>> with pytest.raises(ValueError): ... list(_slices1d(margin, stop, step))