kwimage.structs

mkinit ~/code/kwimage/kwimage/structs/__init__.py -w –relative –nomod

A common thread in many kwimage.structs / kwannot objects is that they attempt to store multiple data elements using a single data structure when possible e.g. the classes are Boxes, Points, Detections, Coords, and not Box, Detection, Coord. The exceptions are Polygon, Heatmap, and Mask, where it made more sense to have one object-per item because each individual item is a reasonably sized chuck of data.

Another commonality is that objects have only two main attributes: .data and .meta. These allow the underlying representation of the object to vary as needed.

Currently Boxes and Mask do not have a .meta attribute. They instead have a .format attribute which is a text-code indicating the underlying layout of the data.

The data and meta instance attributes in the Points, Detections, and Heatmaps classes are dictionaries. These classes also have a __datakeys__ and __metakeys__ class attribute, which are lists of strings. These lists specify which keys are expected in each dictionary. For instance, Points.__datakeys__ = [‘xy’, ‘class_idxs’, ‘visible’] and Points.__metakeys__ = [‘classes’]. All objects in the data dictionary are expected to be aligned, whereas the meta dictionary is for auxillay data. For example in Points, the xy position data[‘xy’][i] is expected to have the class index data[‘class_idxs’][i]. By convention, a class index indexes into the list of category names stored in meta[‘classes’].

The Heatmap.data behaves slighly different than Points. Its data dictionary stores different per-pixel attributes like class probability scores, or offset vectors. The meta dictionary stores data like the originaly image dimensions (heatmaps are usually downsampled wrt the image that they correspond to) and the transformation matrices would warp the “data” space back onto the original image space.

Note that the developer can add any extra data or meta keys that they like, but they should keep in mind that all items in data should be aligned, whereas meta can contain arbitrary information.

Subpackages

• kwimage.structs._boxes_backend
• kwimage.structs._mask_backend

Submodules

• kwimage.structs._generic
• kwimage.structs.boxes
• kwimage.structs.coords
• kwimage.structs.detections
• kwimage.structs.heatmap
• kwimage.structs.mask
• kwimage.structs.points
• kwimage.structs.polygon
• kwimage.structs.segmentation

Package Contents

Classes

Boxes	Converts boxes between different formats as long as the last dimension
Coords	A data structure to store n-dimensional coordinate geometry.
Detections	Container for holding and manipulating multiple detections.
Heatmap	Keeps track of a downscaled heatmap and how to transform it to overlay the
Mask	Manages a single segmentation mask and can convert to and from
MaskList	Store and manipulate multiple masks, usually within the same image
Points	Stores multiple keypoints for a single object.
PointsList	Stores a list of Points, each item usually corresponds to a different object.
MultiPolygon	Data structure for storing multiple polygons (typically related to the same
Polygon	Represents a single polygon as set of exterior boundary points and a list
PolygonList	Stores and allows manipluation of multiple polygons, usually within the
Segmentation	Either holds a MultiPolygon, Polygon, or Mask
SegmentationList	Store and manipulate multiple segmentations (masks or polygons), usually

Functions

smooth_prob(prob, k=3, inplace=False, eps=1e-09)

Smooths the probability map, but preserves the magnitude of the peaks.

class kwimage.structs.Boxes(data, format=None, check=True)

Bases: _BoxConversionMixins, _BoxPropertyMixins, _BoxTransformMixins, _BoxDrawMixins, ubelt.NiceRepr

Converts boxes between different formats as long as the last dimension contains 4 coordinates and the format is specified.

This is a convinience class, and should not not store the data for very long. The general idiom should be create class, convert data, and then get the raw data and let the class be garbage collected. This will help ensure that your code is portable and understandable if this class is not available.

Example

>>> # xdoctest: +IGNORE_WHITESPACE
>>> import kwimage
>>> import numpy as np
>>> # Given an array / tensor that represents one or more boxes
>>> data = np.array([[ 0,  0, 10, 10],
>>>                  [ 5,  5, 50, 50],
>>>                  [20,  0, 30, 10]])
>>> # The kwimage.Boxes data structure is a thin fast wrapper
>>> # that provides methods for operating on the boxes.
>>> # It requires that the user explicitly provide a code that denotes
>>> # the format of the boxes (i.e. what each column represents)
>>> boxes = kwimage.Boxes(data, 'ltrb')
>>> # This means that there is no ambiguity about box format
>>> # The representation string of the Boxes object demonstrates this
>>> print('boxes = {!r}'.format(boxes))
boxes = <Boxes(ltrb,
    array([[ 0,  0, 10, 10],
           [ 5,  5, 50, 50],
           [20,  0, 30, 10]]))>
>>> # if you pass this data around. You can convert to other formats
>>> # For docs on available format codes see :class:`BoxFormat`.
>>> # In this example we will convert (left, top, right, bottom)
>>> # to (left-x, top-y, width, height).
>>> boxes.toformat('xywh')
<Boxes(xywh,
    array([[ 0,  0, 10, 10],
           [ 5,  5, 45, 45],
           [20,  0, 10, 10]]))>
>>> # In addition to format conversion there are other operations
>>> # We can quickly (using a C-backend) find IoUs
>>> ious = boxes.ious(boxes)
>>> print('{}'.format(ub.repr2(ious, nl=1, precision=2, with_dtype=False)))
np.array([[1.  , 0.01, 0.  ],
          [0.01, 1.  , 0.02],
          [0.  , 0.02, 1.  ]])
>>> # We can ask for the area of each box
>>> print('boxes.area = {}'.format(ub.repr2(boxes.area, nl=0, with_dtype=False)))
boxes.area = np.array([[ 100],[2025],[ 100]])
>>> # We can ask for the center of each box
>>> print('boxes.center = {}'.format(ub.repr2(boxes.center, nl=1, with_dtype=False)))
boxes.center = (
    np.array([[ 5. ],[27.5],[25. ]]),
    np.array([[ 5. ],[27.5],[ 5. ]]),
)
>>> # We can translate / scale the boxes
>>> boxes.translate((10, 10)).scale(100)
<Boxes(ltrb,
    array([[1000., 1000., 2000., 2000.],
           [1500., 1500., 6000., 6000.],
           [3000., 1000., 4000., 2000.]]))>
>>> # We can clip the bounding boxes
>>> boxes.translate((10, 10)).scale(100).clip(1200, 1200, 1700, 1800)
<Boxes(ltrb,
    array([[1200., 1200., 1700., 1800.],
           [1500., 1500., 1700., 1800.],
           [1700., 1200., 1700., 1800.]]))>
>>> # We can perform arbitrary warping of the boxes
>>> # (note that if the transform is not axis aligned, the axis aligned
>>> #  bounding box of the transform result will be returned)
>>> transform = np.array([[-0.83907153,  0.54402111,  0. ],
>>>                       [-0.54402111, -0.83907153,  0. ],
>>>                       [ 0.        ,  0.        ,  1. ]])
>>> boxes.warp(transform)
<Boxes(ltrb,
    array([[ -8.3907153 , -13.8309264 ,   5.4402111 ,   0.        ],
           [-39.23347095, -69.154632  ,  23.00569785,  -6.9154632 ],
           [-25.1721459 , -24.7113486 , -11.3412195 , -10.8804222 ]]))>
>>> # Note, that we can transform the box to a Polygon for more
>>> # accurate warping.
>>> transform = np.array([[-0.83907153,  0.54402111,  0. ],
>>>                       [-0.54402111, -0.83907153,  0. ],
>>>                       [ 0.        ,  0.        ,  1. ]])
>>> warped_polys = boxes.to_polygons().warp(transform)
>>> print(ub.repr2(warped_polys.data, sv=1))
[
    <Polygon({
        'exterior': <Coords(data=
                        array([[  0.       ,   0.       ],
                               [  5.4402111,  -8.3907153],
                               [ -2.9505042, -13.8309264],
                               [ -8.3907153,  -5.4402111],
                               [  0.       ,   0.       ]]))>,
        'interiors': [],
    })>,
    <Polygon({
        'exterior': <Coords(data=
                        array([[ -1.4752521 ,  -6.9154632 ],
                               [ 23.00569785, -44.67368205],
                               [-14.752521  , -69.154632  ],
                               [-39.23347095, -31.39641315],
                               [ -1.4752521 ,  -6.9154632 ]]))>,
        'interiors': [],
    })>,
    <Polygon({
        'exterior': <Coords(data=
                        array([[-16.7814306, -10.8804222],
                               [-11.3412195, -19.2711375],
                               [-19.7319348, -24.7113486],
                               [-25.1721459, -16.3206333],
                               [-16.7814306, -10.8804222]]))>,
        'interiors': [],
    })>,
]
>>> # The kwimage.Boxes data structure is also convertable to
>>> # several alternative data structures, like shapely, coco, and imgaug.
>>> print(ub.repr2(boxes.to_shapely(), sv=1))
[
    POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0)),
    POLYGON ((5 5, 5 50, 50 50, 50 5, 5 5)),
    POLYGON ((20 0, 20 10, 30 10, 30 0, 20 0)),
]
>>> # xdoctest: +REQUIRES(module:imgaug)
>>> print(ub.repr2(boxes[0:1].to_imgaug(shape=(100, 100)), sv=1))
BoundingBoxesOnImage([BoundingBox(x1=0.0000, y1=0.0000, x2=10.0000, y2=10.0000, label=None)], shape=(100, 100))
>>> # xdoctest: -REQUIRES(module:imgaug)
>>> print(ub.repr2(list(boxes.to_coco()), sv=1))
[
    [0, 0, 10, 10],
    [5, 5, 45, 45],
    [20, 0, 10, 10],
]
>>> # Finally, when you are done with your boxes object, you can
>>> # unwrap the raw data by using the ``.data`` attribute
>>> # all operations are done on this data, which gives the
>>> # kwiamge.Boxes data structure almost no overhead when
>>> # inserted into existing code.
>>> print('boxes.data =\n{}'.format(ub.repr2(boxes.data, nl=1)))
boxes.data =
np.array([[ 0,  0, 10, 10],
          [ 5,  5, 50, 50],
          [20,  0, 30, 10]], dtype=np.int64)
>>> # xdoctest: +REQUIRES(module:torch)
>>> # This data structure was designed for use with both torch
>>> # and numpy, the underlying data can be either an array or tensor.
>>> boxes.tensor()
<Boxes(ltrb,
    tensor([[ 0,  0, 10, 10],
            [ 5,  5, 50, 50],
            [20,  0, 30, 10]]))>
>>> boxes.numpy()
<Boxes(ltrb,
    array([[ 0,  0, 10, 10],
           [ 5,  5, 50, 50],
           [20,  0, 30, 10]]))>

Example

>>> # xdoctest: +IGNORE_WHITESPACE
>>> from kwimage.structs.boxes import *  # NOQA
>>> # Demo of conversion methods
>>> import kwimage
>>> kwimage.Boxes([[25, 30, 15, 10]], 'xywh')
<Boxes(xywh, array([[25, 30, 15, 10]]))>
>>> kwimage.Boxes([[25, 30, 15, 10]], 'xywh').to_xywh()
<Boxes(xywh, array([[25, 30, 15, 10]]))>
>>> kwimage.Boxes([[25, 30, 15, 10]], 'xywh').to_cxywh()
<Boxes(cxywh, array([[32.5, 35. , 15. , 10. ]]))>
>>> kwimage.Boxes([[25, 30, 15, 10]], 'xywh').to_ltrb()
<Boxes(ltrb, array([[25, 30, 40, 40]]))>
>>> kwimage.Boxes([[25, 30, 15, 10]], 'xywh').scale(2).to_ltrb()
<Boxes(ltrb, array([[50., 60., 80., 80.]]))>
>>> # xdoctest: +REQUIRES(module:torch)
>>> kwimage.Boxes(torch.FloatTensor([[25, 30, 15, 20]]), 'xywh').scale(.1).to_ltrb()
<Boxes(ltrb, tensor([[ 2.5000,  3.0000,  4.0000,  5.0000]]))>

Notes

In the following examples we show cases where Boxes can hold a single 1-dimensional box array. This is a holdover from an older codebase, and some functions may assume that the input is at least 2-D. Thus when representing a single bounding box it is best practice to view it as a list of 1 box. While many function will work in the 1-D case, not all functions have been tested and thus we cannot gaurentee correctness.

Example

>>> # xdoctest: +IGNORE_WHITESPACE
>>> Boxes([25, 30, 15, 10], 'xywh')
<Boxes(xywh, array([25, 30, 15, 10]))>
>>> Boxes([25, 30, 15, 10], 'xywh').to_xywh()
<Boxes(xywh, array([25, 30, 15, 10]))>
>>> Boxes([25, 30, 15, 10], 'xywh').to_cxywh()
<Boxes(cxywh, array([32.5, 35. , 15. , 10. ]))>
>>> Boxes([25, 30, 15, 10], 'xywh').to_ltrb()
<Boxes(ltrb, array([25, 30, 40, 40]))>
>>> Boxes([25, 30, 15, 10], 'xywh').scale(2).to_ltrb()
<Boxes(ltrb, array([50., 60., 80., 80.]))>
>>> # xdoctest: +REQUIRES(module:torch)
>>> Boxes(torch.FloatTensor([[25, 30, 15, 20]]), 'xywh').scale(.1).to_ltrb()
<Boxes(ltrb, tensor([[ 2.5000,  3.0000,  4.0000,  5.0000]]))>

Example

>>> datas = [
>>>     [1, 2, 3, 4],
>>>     [[1, 2, 3, 4], [4, 5, 6, 7]],
>>>     [[[1, 2, 3, 4], [4, 5, 6, 7]]],
>>> ]
>>> formats = BoxFormat.cannonical
>>> for format1 in formats:
>>>     for data in datas:
>>>         self = box1 = Boxes(data, format1)
>>>         for format2 in formats:
>>>             box2 = box1.toformat(format2)
>>>             back = box2.toformat(format1)
>>>             assert box1 == back

__getitem__(self, index)

__eq__(self, other)

Tests equality of two Boxes objects

Example

>>> box0 = box1 = Boxes([[1, 2, 3, 4]], 'xywh')
>>> box2 = Boxes(box0.data, 'ltrb')
>>> box3 = Boxes([[0, 2, 3, 4]], box0.format)
>>> box4 = Boxes(box0.data, box2.format)
>>> assert box0 == box1
>>> assert not box0 == box2
>>> assert not box2 == box3
>>> assert box2 == box4

__len__(self)

__nice__(self)

__repr__(self)

Return repr(self).

classmethod random(Boxes, num=1, scale=1.0, format=BoxFormat.XYWH, anchors=None, anchor_std=1.0 / 6, tensor=False, rng=None)

Makes random boxes; typically for testing purposes

Parameters

• num (int) – number of boxes to generate

• scale (float | Tuple[float, float]) – size of imgdims

• format (str) – format of boxes to be created (e.g. ltrb, xywh)

• anchors (ndarray) – normalized width / heights of anchor boxes to perterb and randomly place. (must be in range 0-1)

• anchor_std (float) – magnitude of noise applied to anchor shapes

• tensor (bool) – if True, returns boxes in tensor format

• rng (None | int | RandomState) – initial random seed

Example

>>> # xdoctest: +IGNORE_WHITESPACE
>>> Boxes.random(3, rng=0, scale=100)
<Boxes(xywh,
    array([[54, 54,  6, 17],
           [42, 64,  1, 25],
           [79, 38, 17, 14]]))>
>>> # xdoctest: +REQUIRES(module:torch)
>>> Boxes.random(3, rng=0, scale=100).tensor()
<Boxes(xywh,
    tensor([[ 54,  54,   6,  17],
            [ 42,  64,   1,  25],
            [ 79,  38,  17,  14]]))>
>>> anchors = np.array([[.5, .5], [.3, .3]])
>>> Boxes.random(3, rng=0, scale=100, anchors=anchors)
<Boxes(xywh,
    array([[ 2, 13, 51, 51],
           [32, 51, 32, 36],
           [36, 28, 23, 26]]))>

Example

>>> # Boxes position/shape within 0-1 space should be uniform.
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> fig = kwplot.figure(fnum=1, doclf=True)
>>> fig.gca().set_xlim(0, 128)
>>> fig.gca().set_ylim(0, 128)
>>> import kwimage
>>> kwimage.Boxes.random(num=10).scale(128).draw()

copy(self)

classmethod concatenate(cls, boxes, axis=0)

Concatenates multiple boxes together

Parameters

• boxes (Sequence[Boxes]) – list of boxes to concatenate

• axis (int, default=0) – axis to stack on

Returns

stacked boxes

Return type

Boxes

Example

>>> boxes = [Boxes.random(3) for _ in range(3)]
>>> new = Boxes.concatenate(boxes)
>>> assert len(new) == 9
>>> assert np.all(new.data[3:6] == boxes[1].data)

Example

>>> boxes = [Boxes.random(3) for _ in range(3)]
>>> boxes[0].data = boxes[0].data[0]
>>> boxes[1].data = boxes[0].data[0:0]
>>> new = Boxes.concatenate(boxes)
>>> assert len(new) == 4
>>> # xdoctest: +REQUIRES(module:torch)
>>> new = Boxes.concatenate([b.tensor() for b in boxes])
>>> assert len(new) == 4

compress(self, flags, axis=0, inplace=False)

Filters boxes based on a boolean criterion

Parameters

• flags (ArrayLike[bool]) – true for items to be kept

• axis (int) – you usually want this to be 0

• inplace (bool) – if True, modifies this object

Example

>>> self = Boxes([[25, 30, 15, 10]], 'ltrb')
>>> self.compress([True])
<Boxes(ltrb, array([[25, 30, 15, 10]]))>
>>> self.compress([False])
<Boxes(ltrb, array([], shape=(0, 4), dtype=int64))>

take(self, idxs, axis=0, inplace=False)

Takes a subset of items at specific indices

Parameters

• indices (ArrayLike[int]) – indexes of items to take

• axis (int) – you usually want this to be 0

• inplace (bool) – if True, modifies this object

Example

>>> self = Boxes([[25, 30, 15, 10]], 'ltrb')
>>> self.take([0])
<Boxes(ltrb, array([[25, 30, 15, 10]]))>
>>> self.take([])
<Boxes(ltrb, array([], shape=(0, 4), dtype=int64))>

is_tensor(self)

is the backend fueled by torch?

is_numpy(self)

is the backend fueled by numpy?

_impl(self)

returns the kwarray.ArrayAPI implementation for the data

Example

>>> assert Boxes.random().numpy()._impl.is_numpy
>>> # xdoctest: +REQUIRES(module:torch)
>>> assert Boxes.random().tensor()._impl.is_tensor

property device(self)

If the backend is torch returns the data device, otherwise None

astype(self, dtype)

Changes the type of the internal array used to represent the boxes

Notes

this operation is not inplace

Example

>>> # xdoctest: +IGNORE_WHITESPACE
>>> # xdoctest: +REQUIRES(module:torch)
>>> Boxes.random(3, 100, rng=0).tensor().astype('int32')
<Boxes(xywh,
    tensor([[54, 54,  6, 17],
            [42, 64,  1, 25],
            [79, 38, 17, 14]], dtype=torch.int32))>
>>> Boxes.random(3, 100, rng=0).numpy().astype('int32')
<Boxes(xywh,
    array([[54, 54,  6, 17],
           [42, 64,  1, 25],
           [79, 38, 17, 14]], dtype=int32))>
>>> Boxes.random(3, 100, rng=0).tensor().astype('float32')
>>> Boxes.random(3, 100, rng=0).numpy().astype('float32')

round(self, inplace=False)

Rounds data coordinates to the nearest integer.

This operation is applied directly to the box coordinates, so its output will depend on the format the boxes are stored in.

Parameters

inplace (bool, default=False) – if True, modifies this object

SeeAlso:

:method:`Boxes.quantize`

Example

>>> import kwimage
>>> self = kwimage.Boxes.random(3, rng=0).scale(10)
>>> new = self.round()
>>> print('self = {!r}'.format(self))
>>> print('new = {!r}'.format(new))
self = <Boxes(xywh,
    array([[5.48813522, 5.44883192, 0.53949833, 1.70306146],
           [4.23654795, 6.4589411 , 0.13932407, 2.45878875],
           [7.91725039, 3.83441508, 1.71937704, 1.45453393]]))>
new = <Boxes(xywh,
    array([[5., 5., 1., 2.],
           [4., 6., 0., 2.],
           [8., 4., 2., 1.]]))>

quantize(self, inplace=False, dtype=np.int32)

Converts the box to integer coordinates.

This operation takes the floor of the left side and the ceil of the right side. Thus the area of the box will never decreases.

Parameters

• inplace (bool, default=False) – if True, modifies this object

• dtype (type) – type to cast as

SeeAlso:

:method:`Boxes.round`

Example

>>> import kwimage
>>> self = kwimage.Boxes.random(3, rng=0).scale(10)
>>> new = self.quantize()
>>> print('self = {!r}'.format(self))
>>> print('new = {!r}'.format(new))
self = <Boxes(xywh,
    array([[5.48813522, 5.44883192, 0.53949833, 1.70306146],
           [4.23654795, 6.4589411 , 0.13932407, 2.45878875],
           [7.91725039, 3.83441508, 1.71937704, 1.45453393]]))>
new = <Boxes(xywh,
    array([[5, 5, 2, 3],
           [4, 6, 1, 3],
           [7, 3, 3, 3]], dtype=int32))>

Example

>>> import kwimage
>>> self = kwimage.Boxes.random(3, rng=0)
>>> orig = self.copy()
>>> self.quantize(inplace=True)
>>> assert np.any(self.data != orig.data)

numpy(self)

Converts tensors to numpy. Does not change memory if possible.

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Boxes.random(3).tensor()
>>> newself = self.numpy()
>>> self.data[0, 0] = 0
>>> assert newself.data[0, 0] == 0
>>> self.data[0, 0] = 1
>>> assert self.data[0, 0] == 1

tensor(self, device=ub.NoParam)

Converts numpy to tensors. Does not change memory if possible.

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Boxes.random(3)
>>> # xdoctest: +REQUIRES(module:torch)
>>> newself = self.tensor()
>>> self.data[0, 0] = 0
>>> assert newself.data[0, 0] == 0
>>> self.data[0, 0] = 1
>>> assert self.data[0, 0] == 1

ious(self, other, bias=0, impl='auto', mode=None)

Intersection over union.

Compute IOUs (intersection area over union area) between these boxes and another set of boxes. This is a symmetric measure of similarity between boxes.

Todo

• [ ] Add pairwise flag to toggle between one-vs-one and all-vs-all

  computation. I.E. Add option for componentwise calculation.

Parameters

• other (Boxes) – boxes to compare IoUs against

• bias (int, default=0) – either 0 or 1, does TL=BR have area of 0 or 1?

• impl (str, default=’auto’) – code to specify implementation used to ious. Can be either torch, py, c, or auto. Efficiency and the exact result will vary by implementation, but they will always be close. Some implementations only accept certain data types (e.g. impl=’c’, only accepts float32 numpy arrays). See ~/code/kwimage/dev/bench_bbox.py for benchmark details. On my system the torch impl was fastest (when the data was on the GPU).

• mode – depricated, use impl

SeeAlso:

iooas - for a measure of coverage between boxes

Examples

>>> import kwimage
>>> self = kwimage.Boxes(np.array([[ 0,  0, 10, 10],
>>>                                [10,  0, 20, 10],
>>>                                [20,  0, 30, 10]]), 'ltrb')
>>> other = kwimage.Boxes(np.array([6, 2, 20, 10]), 'ltrb')
>>> overlaps = self.ious(other, bias=1).round(2)
>>> assert np.all(np.isclose(overlaps, [0.21, 0.63, 0.04])), repr(overlaps)

Examples

>>> import kwimage
>>> boxes1 = kwimage.Boxes(np.array([[ 0,  0, 10, 10],
>>>                                  [10,  0, 20, 10],
>>>                                  [20,  0, 30, 10]]), 'ltrb')
>>> other = kwimage.Boxes(np.array([[6, 2, 20, 10],
>>>                                 [100, 200, 300, 300]]), 'ltrb')
>>> overlaps = boxes1.ious(other)
>>> print('{}'.format(ub.repr2(overlaps, precision=2, nl=1)))
np.array([[0.18, 0.  ],
          [0.61, 0.  ],
          [0.  , 0.  ]]...)

Examples

>>> # xdoctest: +IGNORE_WHITESPACE
>>> Boxes(np.empty(0), 'xywh').ious(Boxes(np.empty(4), 'xywh')).shape
(0,)
>>> #Boxes(np.empty(4), 'xywh').ious(Boxes(np.empty(0), 'xywh')).shape
>>> Boxes(np.empty((0, 4)), 'xywh').ious(Boxes(np.empty((0, 4)), 'xywh')).shape
(0, 0)
>>> Boxes(np.empty((1, 4)), 'xywh').ious(Boxes(np.empty((0, 4)), 'xywh')).shape
(1, 0)
>>> Boxes(np.empty((0, 4)), 'xywh').ious(Boxes(np.empty((1, 4)), 'xywh')).shape
(0, 1)

Examples

>>> # xdoctest: +REQUIRES(module:torch)
>>> formats = BoxFormat.cannonical
>>> istensors = [False, True]
>>> results = {}
>>> for format in formats:
>>>     for tensor in istensors:
>>>         boxes1 = Boxes.random(5, scale=10.0, rng=0, format=format, tensor=tensor)
>>>         boxes2 = Boxes.random(7, scale=10.0, rng=1, format=format, tensor=tensor)
>>>         ious = boxes1.ious(boxes2)
>>>         results[(format, tensor)] = ious
>>> results = {k: v.numpy() if torch.is_tensor(v) else v for k, v in results.items() }
>>> results = {k: v.tolist() for k, v in results.items()}
>>> print(ub.repr2(results, sk=True, precision=3, nl=2))
>>> from functools import partial
>>> assert ub.allsame(results.values(), partial(np.allclose, atol=1e-07))

Ignore:

>>> # does this work with backprop?
>>> # xdoctest: +REQUIRES(module:torch)
>>> import torch
>>> import kwimage
>>> num = 1000
>>> true_boxes = kwimage.Boxes.random(num).tensor()
>>> inputs = torch.rand(num, 10)
>>> regress = torch.nn.Linear(10, 4)
>>> energy = regress(inputs)
>>> energy.retain_grad()
>>> outputs = energy.sigmoid()
>>> outputs.retain_grad()
>>> out_boxes = kwimage.Boxes(outputs, 'cxywh')
>>> ious = out_boxes.ious(true_boxes)
>>> loss = ious.sum()
>>> loss.backward()

iooas(self, other, bias=0)

Intersection over other area.

This is an asymetric measure of coverage. How much of the “other” boxes are covered by these boxes. It is the area of intersection between each pair of boxes and the area of the “other” boxes.

SeeAlso:

ious - for a measure of similarity between boxes

Parameters

• other (Boxes) – boxes to compare IoOA against

• bias (int, default=0) – either 0 or 1, does TL=BR have area of 0 or 1?

Examples

>>> self = Boxes(np.array([[ 0,  0, 10, 10],
>>>                        [10,  0, 20, 10],
>>>                        [20,  0, 30, 10]]), 'ltrb')
>>> other = Boxes(np.array([[6, 2, 20, 10], [0, 0, 0, 3]]), 'xywh')
>>> coverage = self.iooas(other, bias=0).round(2)
>>> print('coverage = {!r}'.format(coverage))

isect_area(self, other, bias=0)

Intersection part of intersection over union computation

Examples

>>> # xdoctest: +IGNORE_WHITESPACE
>>> self = Boxes.random(5, scale=10.0, rng=0, format='ltrb')
>>> other = Boxes.random(3, scale=10.0, rng=1, format='ltrb')
>>> isect = self.isect_area(other, bias=0)
>>> ious_v1 = isect / ((self.area + other.area.T) - isect)
>>> ious_v2 = self.ious(other, bias=0)
>>> assert np.allclose(ious_v1, ious_v2)

intersection(self, other)

Componentwise intersection between two sets of Boxes

intersections of boxes are always boxes, so this works

Returns

intersected boxes

Return type

Boxes

Examples

>>> # xdoctest: +IGNORE_WHITESPACE
>>> from kwimage.structs.boxes import *  # NOQA
>>> self = Boxes.random(5, rng=0).scale(10.)
>>> other = self.translate(1)
>>> new = self.intersection(other)
>>> new_area = np.nan_to_num(new.area).ravel()
>>> alt_area = np.diag(self.isect_area(other))
>>> close = np.isclose(new_area, alt_area)
>>> assert np.all(close)

union_hull(self, other)

Componentwise hull union between two sets of Boxes

NOTE: convert to polygon to do a real union.

Returns

unioned boxes

Return type

Boxes

Examples

>>> # xdoctest: +IGNORE_WHITESPACE
>>> from kwimage.structs.boxes import *  # NOQA
>>> self = Boxes.random(5, rng=0).scale(10.)
>>> other = self.translate(1)
>>> new = self.union_hull(other)
>>> new_area = np.nan_to_num(new.area).ravel()

bounding_box(self)

Returns the box that bounds all of the contained boxes

Returns

a single box

Return type

Boxes

Examples

>>> # xdoctest: +IGNORE_WHITESPACE
>>> from kwimage.structs.boxes import *  # NOQA
>>> self = Boxes.random(5, rng=0).scale(10.)
>>> other = self.translate(1)
>>> new = self.union_hull(other)
>>> new_area = np.nan_to_num(new.area).ravel()

contains(self, other)

Determine of points are completely contained by these boxes

Parameters

other (Points) – points to test for containment. TODO: support generic data types

Returns

N x M boolean matrix indicating which box

contains which points, where N is the number of boxes and M is the number of points.

Return type

flags (ArrayLike)

Examples

>>> import kwimage
>>> self = kwimage.Boxes.random(10).scale(10).round()
>>> other = kwimage.Points.random(10).scale(10).round()
>>> flags = self.contains(other)
>>> flags = self.contains(self.xy_center)
>>> assert np.all(np.diag(flags))

view(self, *shape)

Passthrough method to view or reshape

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Boxes.random(6, scale=10.0, rng=0, format='xywh').tensor()
>>> assert list(self.view(3, 2, 4).data.shape) == [3, 2, 4]
>>> self = Boxes.random(6, scale=10.0, rng=0, format='ltrb').tensor()
>>> assert list(self.view(3, 2, 4).data.shape) == [3, 2, 4]

class kwimage.structs.Coords(data=None, meta=None)

Bases: kwimage.structs._generic.Spatial, ubelt.NiceRepr

A data structure to store n-dimensional coordinate geometry.

Currently it is up to the user to maintain what coordinate system this geometry belongs to.

Note

This class was designed to hold coordinates in r/c format, but in general this class is anostic to dimension ordering as long as you are consistent. However, there are two places where this matters:

(1) drawing and (2) gdal/imgaug-warping. In these places we will assume x/y for legacy reasons. This may change in the future.

The term axes with resepct to Coords always refers to the final numpy axis. In other words the final numpy-axis represents ALL of the coordinate-axes.

CommandLine:

xdoctest -m kwimage.structs.coords Coords

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> import kwarray
>>> rng = kwarray.ensure_rng(0)
>>> self = Coords.random(num=4, dim=3, rng=rng)
>>> print('self = {}'.format(self))
self = <Coords(data=
    array([[0.5488135 , 0.71518937, 0.60276338],
           [0.54488318, 0.4236548 , 0.64589411],
           [0.43758721, 0.891773  , 0.96366276],
           [0.38344152, 0.79172504, 0.52889492]]))>
>>> matrix = rng.rand(4, 4)
>>> self.warp(matrix)
<Coords(data=
    array([[0.71037426, 1.25229659, 1.39498435],
           [0.60799503, 1.26483447, 1.42073131],
           [0.72106004, 1.39057144, 1.38757508],
           [0.68384299, 1.23914654, 1.29258196]]))>
>>> self.translate(3, inplace=True)
<Coords(data=
    array([[3.5488135 , 3.71518937, 3.60276338],
           [3.54488318, 3.4236548 , 3.64589411],
           [3.43758721, 3.891773  , 3.96366276],
           [3.38344152, 3.79172504, 3.52889492]]))>
>>> self.translate(3, inplace=True)
<Coords(data=
    array([[6.5488135 , 6.71518937, 6.60276338],
           [6.54488318, 6.4236548 , 6.64589411],
           [6.43758721, 6.891773  , 6.96366276],
           [6.38344152, 6.79172504, 6.52889492]]))>
>>> self.scale(2)
<Coords(data=
    array([[13.09762701, 13.43037873, 13.20552675],
           [13.08976637, 12.8473096 , 13.29178823],
           [12.87517442, 13.783546  , 13.92732552],
           [12.76688304, 13.58345008, 13.05778984]]))>
>>> # xdoctest: +REQUIRES(module:torch)
>>> self.tensor()
>>> self.tensor().tensor().numpy().numpy()
>>> self.numpy()
>>> #self.draw_on()

__repr__

__nice__(self)

__len__(self)

property dtype(self)

property dim(self)

property shape(self)

copy(self)

classmethod random(Coords, num=1, dim=2, rng=None, meta=None)

Makes random coordinates; typically for testing purposes

is_numpy(self)

is_tensor(self)

compress(self, flags, axis=0, inplace=False)

Filters items based on a boolean criterion

Parameters

• flags (ArrayLike[bool]) – true for items to be kept

• axis (int) – you usually want this to be 0

• inplace (bool, default=False) – if True, modifies this object

Returns

filtered coords

Return type

Coords

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, rng=0)
>>> self.compress([True] * len(self))
>>> self.compress([False] * len(self))
<Coords(data=array([], shape=(0, 2), dtype=float64))>
>>> # xdoctest: +REQUIRES(module:torch)
>>> self = self.tensor()
>>> self.compress([True] * len(self))
>>> self.compress([False] * len(self))

take(self, indices, axis=0, inplace=False)

Takes a subset of items at specific indices

Parameters

• indices (ArrayLike[int]) – indexes of items to take

• axis (int) – you usually want this to be 0

• inplace (bool, default=False) – if True, modifies this object

Returns

filtered coords

Return type

Coords

Example

>>> self = Coords(np.array([[25, 30, 15, 10]]))
>>> self.take([0])
<Coords(data=array([[25, 30, 15, 10]]))>
>>> self.take([])
<Coords(data=array([], shape=(0, 4), dtype=int64))>

astype(self, dtype, inplace=False)

Changes the data type

Parameters

• dtype – new type

• inplace (bool, default=False) – if True, modifies this object

Returns

modified coordinates

Return type

Coords

round(self, inplace=False)

Rounds data to the nearest integer

Parameters

inplace (bool, default=False) – if True, modifies this object

Example

>>> import kwimage
>>> self = kwimage.Coords.random(3).scale(10)
>>> self.round()

view(self, *shape)

Passthrough method to view or reshape

Parameters

*shape – new shape of the data

Returns

modified coordinates

Return type

Coords

Example

>>> self = Coords.random(6, dim=4).numpy()
>>> assert list(self.view(3, 2, 4).data.shape) == [3, 2, 4]
>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Coords.random(6, dim=4).tensor()
>>> assert list(self.view(3, 2, 4).data.shape) == [3, 2, 4]

classmethod concatenate(cls, coords, axis=0)

Concatenates lists of coordinates together

Parameters

• coords (Sequence[Coords]) – list of coords to concatenate

• axis (int, default=0) – axis to stack on

Returns

stacked coords

Return type

Coords

CommandLine:

xdoctest -m kwimage.structs.coords Coords.concatenate

Example

>>> coords = [Coords.random(3) for _ in range(3)]
>>> new = Coords.concatenate(coords)
>>> assert len(new) == 9
>>> assert np.all(new.data[3:6] == coords[1].data)

property device(self)

If the backend is torch returns the data device, otherwise None

property _impl(self)

Returns the internal tensor/numpy ArrayAPI implementation

tensor(self, device=ub.NoParam)

Converts numpy to tensors. Does not change memory if possible.

Returns

modified coordinates

Return type

Coords

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Coords.random(3).numpy()
>>> newself = self.tensor()
>>> self.data[0, 0] = 0
>>> assert newself.data[0, 0] == 0
>>> self.data[0, 0] = 1
>>> assert self.data[0, 0] == 1

numpy(self)

Converts tensors to numpy. Does not change memory if possible.

Returns

modified coordinates

Return type

Coords

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Coords.random(3).tensor()
>>> newself = self.numpy()
>>> self.data[0, 0] = 0
>>> assert newself.data[0, 0] == 0
>>> self.data[0, 0] = 1
>>> assert self.data[0, 0] == 1

reorder_axes(self, new_order, inplace=False)

Change the ordering of the coordinate axes.

Parameters

• new_order (Tuple[int]) – new_order[i] should specify which axes in the original coordinates should be mapped to the i-th position in the returned axes.

• inplace (bool, default=False) – if True, modifies data inplace

Returns

modified coordinates

Return type

Coords

Note

This is the ordering of the “columns” in final numpy axis, not the numpy axes themselves.

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords(data=np.array([
>>>     [7, 11],
>>>     [13, 17],
>>>     [21, 23],
>>> ]))
>>> new = self.reorder_axes((1, 0))
>>> print('new = {!r}'.format(new))
new = <Coords(data=
    array([[11,  7],
           [17, 13],
           [23, 21]]))>

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, rng=0)
>>> new = self.reorder_axes((1, 0))
>>> # Remapping using 1, 0 reverses the axes
>>> assert np.all(new.data[:, 0] == self.data[:, 1])
>>> assert np.all(new.data[:, 1] == self.data[:, 0])
>>> # Remapping using 0, 1 does nothing
>>> eye = self.reorder_axes((0, 1))
>>> assert np.all(eye.data == self.data)
>>> # Remapping using 0, 0, destroys the 1-th column
>>> bad = self.reorder_axes((0, 0))
>>> assert np.all(bad.data[:, 0] == self.data[:, 0])
>>> assert np.all(bad.data[:, 1] == self.data[:, 0])

warp(self, transform, input_dims=None, output_dims=None, inplace=False)

Generalized coordinate transform.

Parameters

• transform (GeometricTransform | ArrayLike | Augmenter | callable) – scikit-image tranform, a 3x3 transformation matrix, an imgaug Augmenter, or generic callable which transforms an NxD ndarray.

• input_dims (Tuple) – shape of the image these objects correspond to (only needed / used when transform is an imgaug augmenter)

• output_dims (Tuple) – unused in non-raster structures, only exists for compatibility.

• inplace (bool, default=False) – if True, modifies data inplace

Returns

modified coordinates

Return type

Coords

Notes

Let D = self.dims

transformation matrices can be either:

• (D + 1) x (D + 1) # for homog

• D x D # for scale / rotate

• D x (D + 1) # for affine

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, rng=0)
>>> transform = skimage.transform.AffineTransform(scale=(2, 2))
>>> new = self.warp(transform)
>>> assert np.all(new.data == self.scale(2).data)

Doctest:

>>> self = Coords.random(10, rng=0)
>>> assert np.all(self.warp(np.eye(3)).data == self.data)
>>> assert np.all(self.warp(np.eye(2)).data == self.data)

Doctest:

>>> # xdoctest: +REQUIRES(module:osgeo)
>>> from osgeo import osr
>>> wgs84_crs = osr.SpatialReference()
>>> wgs84_crs.ImportFromEPSG(4326)
>>> dst_crs = osr.SpatialReference()
>>> dst_crs.ImportFromEPSG(2927)
>>> transform = osr.CoordinateTransformation(wgs84_crs, dst_crs)
>>> self = Coords.random(10, rng=0)
>>> new = self.warp(transform)
>>> assert np.all(new.data != self.data)

>>> # Alternative using generic func
>>> def _gdal_coord_tranform(pts):
...     return np.array([transform.TransformPoint(x, y, 0)[0:2]
...                      for x, y in pts])
>>> alt = self.warp(_gdal_coord_tranform)
>>> assert np.all(alt.data != self.data)
>>> assert np.all(alt.data == new.data)

Doctest:

>>> # can use a generic function
>>> def func(xy):
...     return np.zeros_like(xy)
>>> self = Coords.random(10, rng=0)
>>> assert np.all(self.warp(func).data == 0)

_warp_imgaug(self, augmenter, input_dims, inplace=False)

Warps by applying an augmenter from the imgaug library

Note

We are assuming you are using X/Y coordinates here.

Parameters

• augmenter (imgaug.augmenters.Augmenter)

• input_dims (Tuple) – h/w of the input image

• inplace (bool, default=False) – if True, modifies data inplace

CommandLine:

xdoctest -m ~/code/kwimage/kwimage/structs/coords.py Coords._warp_imgaug

Example

>>> # xdoctest: +REQUIRES(module:imgaug)
>>> from kwimage.structs.coords import *  # NOQA
>>> import imgaug
>>> input_dims = (10, 10)
>>> self = Coords.random(10).scale(input_dims)
>>> augmenter = imgaug.augmenters.Fliplr(p=1)
>>> new = self._warp_imgaug(augmenter, input_dims)
>>> # y coordinate should not change
>>> assert np.allclose(self.data[:, 1], new.data[:, 1])
>>> assert np.allclose(input_dims[0] - self.data[:, 0], new.data[:, 0])

>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> from matplotlib import pyplot as pl
>>> ax = plt.gca()
>>> ax.set_xlim(0, input_dims[0])
>>> ax.set_ylim(0, input_dims[1])
>>> self.draw(color='red', alpha=.4, radius=0.1)
>>> new.draw(color='blue', alpha=.4, radius=0.1)

Example

>>> # xdoctest: +REQUIRES(module:imgaug)
>>> from kwimage.structs.coords import *  # NOQA
>>> import imgaug
>>> input_dims = (32, 32)
>>> inplace = 0
>>> self = Coords.random(1000, rng=142).scale(input_dims).scale(.8)
>>> self.data = self.data.astype(np.int32).astype(np.float32)
>>> augmenter = imgaug.augmenters.CropAndPad(px=(-4, 4), keep_size=1).to_deterministic()
>>> new = self._warp_imgaug(augmenter, input_dims)
>>> # Change should be linear
>>> norm1 = (self.data - self.data.min(axis=0)) / (self.data.max(axis=0) - self.data.min(axis=0))
>>> norm2 = (new.data - new.data.min(axis=0)) / (new.data.max(axis=0) - new.data.min(axis=0))
>>> diff = norm1 - norm2
>>> assert np.allclose(diff, 0, atol=1e-6, rtol=1e-4)
>>> #assert np.allclose(self.data[:, 1], new.data[:, 1])
>>> #assert np.allclose(input_dims[0] - self.data[:, 0], new.data[:, 0])
>>> # xdoc: +REQUIRES(--show)
>>> import kwimage
>>> im = kwimage.imresize(kwimage.grab_test_image(), dsize=input_dims[::-1])
>>> new_im = augmenter.augment_image(im)
>>> import kwplot
>>> plt = kwplot.autoplt()
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.imshow(im, pnum=(1, 2, 1), fnum=1)
>>> self.draw(color='red', alpha=.8, radius=0.5)
>>> kwplot.imshow(new_im, pnum=(1, 2, 2), fnum=1)
>>> new.draw(color='blue', alpha=.8, radius=0.5, coord_axes=[1, 0])

to_imgaug(self, input_dims)

Translate to an imgaug object

Returns

imgaug data structure

Return type

imgaug.KeypointsOnImage

Example

>>> # xdoctest: +REQUIRES(module:imgaug)
>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10)
>>> input_dims = (10, 10)
>>> kpoi = self.to_imgaug(input_dims)
>>> new = Coords.from_imgaug(kpoi)
>>> assert np.allclose(new.data, self.data)

classmethod from_imgaug(cls, kpoi)

scale(self, factor, about=None, output_dims=None, inplace=False)

Scale coordinates by a factor

Parameters

• factor (float or Tuple[float, float]) – scale factor as either a scalar or per-dimension tuple.

• about (Tuple | None) – if unspecified scales about the origin (0, 0), otherwise the rotation is about this point.

• output_dims (Tuple) – unused in non-raster spatial structures

• inplace (bool, default=False) – if True, modifies data inplace

Returns

modified coordinates

Return type

Coords

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, rng=0)
>>> new = self.scale(10)
>>> assert new.data.max() <= 10

>>> self = Coords.random(10, rng=0)
>>> self.data = (self.data * 10).astype(int)
>>> new = self.scale(10)
>>> assert new.data.dtype.kind == 'i'
>>> new = self.scale(10.0)
>>> assert new.data.dtype.kind == 'f'

translate(self, offset, output_dims=None, inplace=False)

Shift the coordinates

Parameters

• offset (float or Tuple[float]) – transation offset as either a scalar or a per-dimension tuple.

• output_dims (Tuple) – unused in non-raster spatial structures

• inplace (bool, default=False) – if True, modifies data inplace

Returns

modified coordinates

Return type

Coords

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, dim=3, rng=0)
>>> new = self.translate(10)
>>> assert new.data.min() >= 10
>>> assert new.data.max() <= 11
>>> Coords.random(3, dim=3, rng=0)
>>> Coords.random(3, dim=3, rng=0).translate((1, 2, 3))

rotate(self, theta, about=None, output_dims=None, inplace=False)

Rotate the coordinates about a point.

Parameters

• theta (float) – rotation angle in radians

• about (Tuple | None) – if unspecified rotates about the origin (0, 0), otherwise the rotation is about this point.

• output_dims (Tuple) – unused in non-raster spatial structures

• inplace (bool, default=False) – if True, modifies data inplace

Returns

modified coordinates

Return type

Coords

Todo

• [ ] Generalized ND Rotations?

References

https://math.stackexchange.com/questions/197772/gen-rot-matrix

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, dim=2, rng=0)
>>> theta = np.pi / 2
>>> new = self.rotate(theta)

>>> # Test rotate agrees with warp
>>> sin_ = np.sin(theta)
>>> cos_ = np.cos(theta)
>>> rot_ = np.array([[cos_, -sin_], [sin_,  cos_]])
>>> new2 = self.warp(rot_)
>>> assert np.allclose(new.data, new2.data)

>>> #
>>> # Rotate about a custom point
>>> theta = np.pi / 2
>>> new3 = self.rotate(theta, about=(0.5, 0.5))
>>> #
>>> # Rotate about the center of mass
>>> about = self.data.mean(axis=0)
>>> new4 = self.rotate(theta, about=about)
>>> # xdoc: +REQUIRES(--show)
>>> # xdoc: +REQUIRES(module:kwplot)
>>> import kwplot
>>> kwplot.figure(fnum=1, doclf=True)
>>> plt = kwplot.autoplt()
>>> self.draw(radius=0.01, color='blue', alpha=.5, coord_axes=[1, 0], setlim='grow')
>>> plt.gca().set_aspect('equal')
>>> new3.draw(radius=0.01, color='red', alpha=.5, coord_axes=[1, 0], setlim='grow')

_rectify_about(self, about)

Ensures that about returns a specified point. Allows for special keys like center to be used.

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10, dim=2, rng=0)

fill(self, image, value, coord_axes=None, interp='bilinear')

Sets sub-coordinate locations in a grid to a particular value

Parameters

coord_axes (Tuple) – specify which image axes each coordinate dim corresponds to. For 2D images, if you are storing r/c data, set to [0,1], if you are storing x/y data, set to [1,0].

Returns

image with coordinates rasterized on it

Return type

ndarray

soft_fill(self, image, coord_axes=None, radius=5)

Used for drawing keypoint truth in heatmaps

Parameters

coord_axes (Tuple) – specify which image axes each coordinate dim corresponds to. For 2D images, if you are storing r/c data, set to [0,1], if you are storing x/y data, set to [1,0].

In other words the i-th entry in coord_axes specifies which row-major spatial dimension the i-th column of a coordinate corresponds to. The index is the coordinate dimension and the value is the axes dimension.

Returns

image with coordinates rasterized on it

Return type

ndarray

References

https://stackoverflow.com/questions/54726703/generating-keypoint-heatmaps-in-tensorflow

Example

>>> from kwimage.structs.coords import *  # NOQA
>>> s = 64
>>> self = Coords.random(10, meta={'shape': (s, s)}).scale(s)
>>> # Put points on edges to to verify "edge cases"
>>> self.data[1] = [0, 0]       # top left
>>> self.data[2] = [s, s]       # bottom right
>>> self.data[3] = [0, s + 10]  # bottom left
>>> self.data[4] = [-3, s // 2] # middle left
>>> self.data[5] = [s + 1, -1]  # top right
>>> # Put points in the middle to verify overlap blending
>>> self.data[6] = [32.5, 32.5] # middle
>>> self.data[7] = [34.5, 34.5] # middle
>>> fill_value = 1
>>> coord_axes = [1, 0]
>>> radius = 10
>>> image1 = np.zeros((s, s))
>>> self.soft_fill(image1, coord_axes=coord_axes, radius=radius)
>>> radius = 3.0
>>> image2 = np.zeros((s, s))
>>> self.soft_fill(image2, coord_axes=coord_axes, radius=radius)
>>> # xdoc: +REQUIRES(--show)
>>> # xdoc: +REQUIRES(module:kwplot)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.imshow(image1, pnum=(1, 2, 1))
>>> kwplot.imshow(image2, pnum=(1, 2, 2))

draw_on(self, image=None, fill_value=1, coord_axes=[1, 0], interp='bilinear')

Note

unlike other methods, the defaults assume x/y internal data

Parameters

coord_axes (Tuple) – specify which image axes each coordinate dim corresponds to. For 2D images, if you are storing r/c data, set to [0,1], if you are storing x/y data, set to [1,0].

In other words the i-th entry in coord_axes specifies which row-major spatial dimension the i-th column of a coordinate corresponds to. The index is the coordinate dimension and the value is the axes dimension.

Returns

image with coordinates drawn on it

Return type

ndarray

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.coords import *  # NOQA
>>> s = 256
>>> self = Coords.random(10, meta={'shape': (s, s)}).scale(s)
>>> self.data[0] = [10, 10]
>>> self.data[1] = [20, 40]
>>> image = np.zeros((s, s))
>>> fill_value = 1
>>> image = self.draw_on(image, fill_value, coord_axes=[1, 0], interp='bilinear')
>>> # image = self.draw_on(image, fill_value, coord_axes=[0, 1], interp='nearest')
>>> # image = self.draw_on(image, fill_value, coord_axes=[1, 0], interp='bilinear')
>>> # image = self.draw_on(image, fill_value, coord_axes=[1, 0], interp='nearest')
>>> # xdoc: +REQUIRES(--show)
>>> # xdoc: +REQUIRES(module:kwplot)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.imshow(image)
>>> self.draw(radius=3, alpha=.5, coord_axes=[1, 0])

draw(self, color='blue', ax=None, alpha=None, coord_axes=[1, 0], radius=1, setlim=False)

Note

unlike other methods, the defaults assume x/y internal data

Parameters

• setlim (bool) – if True ensures the limits of the axes contains the polygon

• coord_axes (Tuple) – specify which image axes each coordinate dim corresponds to. For 2D images,

  if you are storing r/c data, set to [0,1], if you are storing x/y data, set to [1,0].

Returns

drawn matplotlib objects

Return type

List[mpl.collections.PatchCollection]

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.coords import *  # NOQA
>>> self = Coords.random(10)
>>> # xdoc: +REQUIRES(--show)
>>> self.draw(radius=3.0, setlim=True)
>>> import kwplot
>>> kwplot.autompl()
>>> self.draw(radius=3.0)

class kwimage.structs.Detections(data=None, meta=None, datakeys=None, metakeys=None, checks=True, **kwargs)

Bases: ubelt.NiceRepr, _DetAlgoMixin, _DetDrawMixin

Container for holding and manipulating multiple detections.

Variables

• data (Dict) –

  dictionary containing corresponding lists. The length of each list is the number of detections. This contains the bounding boxes, confidence scores, and class indices. Details of the most common keys and types are as follows:

  boxes (kwimage.Boxes[ArrayLike]): multiple bounding boxes scores (ArrayLike): associated scores class_idxs (ArrayLike): associated class indices segmentations (ArrayLike): segmentations masks for each box,

  members can be Mask or MultiPolygon.

  keypoints (ArrayLike): keypoints for each box. Members should

  be Points.

  Additional custom keys may be specified as long as (a) the values are array-like and the first axis corresponds to the standard data values and (b) are custom keys are listed in the datakeys kwargs when constructing the Detections.

• meta (Dict) – This contains contextual information about the detections. This includes the class names, which can be indexed into via the class indexes.

Example

>>> import kwimage
>>> dets = kwimage.Detections(
>>>     # there are expected keys that do not need registration
>>>     boxes=kwimage.Boxes.random(3),
>>>     class_idxs=[0, 1, 1],
>>>     classes=['a', 'b'],
>>>     # custom data attrs must align with boxes
>>>     myattr1=np.random.rand(3),
>>>     myattr2=np.random.rand(3, 2, 8),
>>>     # there are no restrictions on metadata
>>>     mymeta='a custom metadata string',
>>>     # Note that any key not in kwimage.Detections.__datakeys__ or
>>>     # kwimage.Detections.__metakeys__ must be registered at the
>>>     # time of construction.
>>>     datakeys=['myattr1', 'myattr2'],
>>>     metakeys=['mymeta'],
>>>     checks=True,
>>> )
>>> print('dets = {}'.format(dets))
dets = <Detections(3)>

__datakeys__ = ['boxes', 'scores', 'class_idxs', 'probs', 'weights', 'keypoints', 'segmentations']

__metakeys__ = ['classes']

__nice__(self)

__len__(self)

copy(self)

Returns a deep copy of this Detections object

classmethod coerce(cls, data=None, **kwargs)

The “try-anything to get what I want” constructor

Parameters

• data

• **kwargs – currently boxes and cnames

Example

>>> from kwimage.structs.detections import *  # NOQA
>>> import kwimage
>>> kwargs = dict(
>>>     boxes=kwimage.Boxes.random(4),
>>>     cnames=['a', 'b', 'c', 'c'],
>>> )
>>> data = {}
>>> self = kwimage.Detections.coerce(data, **kwargs)

classmethod from_coco_annots(cls, anns, cats=None, classes=None, kp_classes=None, shape=None, dset=None)

Create a Detections object from a list of coco-like annotations.

Parameters

• anns (List[Dict]) – list of coco-like annotation objects

• dset (CocoDataset) – if specified, cats, classes, and kp_classes can are ignored.

• cats (List[Dict]) – coco-format category information. Used only if dset is not specified.

• classes (ndsampler.CategoryTree) – category tree with coco class info. Used only if dset is not specified.

• kp_classes (ndsampler.CategoryTree) – keypoint category tree with coco keypoint class info. Used only if dset is not specified.

• shape (tuple) – shape of parent image

Returns

a detections object

Return type

Detections

Example

>>> from kwimage.structs.detections import *  # NOQA
>>> # xdoctest: +REQUIRES(--module:ndsampler)
>>> anns = [{
>>>     'id': 0,
>>>     'image_id': 1,
>>>     'category_id': 2,
>>>     'bbox': [2, 3, 10, 10],
>>>     'keypoints': [4.5, 4.5, 2],
>>>     'segmentation': {
>>>         'counts': '_11a04M2O0O20N101N3L_5',
>>>         'size': [20, 20],
>>>     },
>>> }]
>>> dataset = {
>>>     'images': [],
>>>     'annotations': [],
>>>     'categories': [
>>>         {'id': 0, 'name': 'background'},
>>>         {'id': 2, 'name': 'class1', 'keypoints': ['spot']}
>>>     ]
>>> }
>>> #import ndsampler
>>> #dset = ndsampler.CocoDataset(dataset)
>>> cats = dataset['categories']
>>> dets = Detections.from_coco_annots(anns, cats)

Example

>>> # xdoctest: +REQUIRES(--module:ndsampler)
>>> # Test case with no category information
>>> from kwimage.structs.detections import *  # NOQA
>>> anns = [{
>>>     'id': 0,
>>>     'image_id': 1,
>>>     'category_id': None,
>>>     'bbox': [2, 3, 10, 10],
>>>     'prob': [.1, .9],
>>> }]
>>> cats = [
>>>     {'id': 0, 'name': 'background'},
>>>     {'id': 2, 'name': 'class1'}
>>> ]
>>> dets = Detections.from_coco_annots(anns, cats)

Example

>>> import kwimage
>>> # xdoctest: +REQUIRES(--module:ndsampler)
>>> import ndsampler
>>> sampler = ndsampler.CocoSampler.demo('photos')
>>> iminfo, anns = sampler.load_image_with_annots(1)
>>> shape = iminfo['imdata'].shape[0:2]
>>> kp_classes = sampler.dset.keypoint_categories()
>>> dets = kwimage.Detections.from_coco_annots(
>>>     anns, sampler.dset.dataset['categories'], sampler.catgraph,
>>>     kp_classes, shape=shape)

to_coco(self, cname_to_cat=None, style='orig', image_id=None, dset=None)

Converts this set of detections into coco-like annotation dictionaries.

Notes

Not all aspects of the MS-COCO format can be accurately represented, so some liberties are taken. The MS-COCO standard defines that annotations should specifiy a category_id field, but in some cases this information is not available so we will populate a ‘category_name’ field if possible and in the worst case fall back to ‘category_index’.

Additionally, detections may contain additional information beyond the MS-COCO standard, and this information (e.g. weight, prob, score) is added as forign fields.

Parameters

• cname_to_cat – currently ignored.

• style (str, default=’orig’) – either ‘orig’ (for the original coco format) or ‘new’ for the more general kwcoco-style coco format.

• image_id (int, default=None) – if specified, populates the image_id field of each image

• dset (CocoDataset, default=None) – if specified, attempts to populate the category_id field to be compatible with this coco dataset.

Yields

dict – coco-like annotation structures

Example

>>> # xdoctest: +REQUIRES(module:ndsampler)
>>> from kwimage.structs.detections import *
>>> self = Detections.demo()[0]
>>> cname_to_cat = None
>>> list(self.to_coco())

property boxes(self)

property class_idxs(self)

property scores(self)

typically only populated for predicted detections

property probs(self)

typically only populated for predicted detections

property weights(self)

typically only populated for groundtruth detections

property classes(self)

num_boxes(self)

warp(self, transform, input_dims=None, output_dims=None, inplace=False)

Spatially warp the detections.

Example

>>> import skimage
>>> transform = skimage.transform.AffineTransform(scale=(2, 3), translation=(4, 5))
>>> self = Detections.random(2)
>>> new = self.warp(transform)
>>> assert new.boxes == self.boxes.warp(transform)
>>> assert new != self

scale(self, factor, output_dims=None, inplace=False)

Spatially warp the detections.

Example

>>> import skimage
>>> transform = skimage.transform.AffineTransform(scale=(2, 3), translation=(4, 5))
>>> self = Detections.random(2)
>>> new = self.warp(transform)
>>> assert new.boxes == self.boxes.warp(transform)
>>> assert new != self

translate(self, offset, output_dims=None, inplace=False)

Spatially warp the detections.

Example

>>> import skimage
>>> self = Detections.random(2)
>>> new = self.translate(10)

classmethod concatenate(cls, dets)

Parameters

boxes (Sequence[Detections]) – list of detections to concatenate

Returns

stacked detections

Return type

Detections

Example

>>> self = Detections.random(2)
>>> other = Detections.random(3)
>>> dets = [self, other]
>>> new = Detections.concatenate(dets)
>>> assert new.num_boxes() == 5

>>> self = Detections.random(2, segmentations=True)
>>> other = Detections.random(3, segmentations=True)
>>> dets = [self, other]
>>> new = Detections.concatenate(dets)
>>> assert new.num_boxes() == 5

argsort(self, reverse=True)

Sorts detection indices by descending (or ascending) scores

Returns

sorted indices

Return type

ndarray[int]

sort(self, reverse=True)

Sorts detections by descending (or ascending) scores

Returns

sorted copy of self

Return type

kwimage.structs.Detections

compress(self, flags, axis=0)

Returns a subset where corresponding locations are True.

Parameters

flags (ndarray[bool]) – mask marking selected items

Returns

subset of self

Return type

kwimage.structs.Detections

CommandLine:

xdoctest -m kwimage.structs.detections Detections.compress

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> import kwimage
>>> dets = kwimage.Detections.random(keypoints='dense')
>>> flags = np.random.rand(len(dets)) > 0.5
>>> subset = dets.compress(flags)
>>> assert len(subset) == flags.sum()
>>> subset = dets.tensor().compress(flags)
>>> assert len(subset) == flags.sum()

take(self, indices, axis=0)

Returns a subset specified by indices

Parameters

indices (ndarray[int]) – indices to select

Returns

subset of self

Return type

kwimage.structs.Detections

Example

>>> import kwimage
>>> dets = kwimage.Detections(boxes=kwimage.Boxes.random(10))
>>> subset = dets.take([2, 3, 5, 7])
>>> assert len(subset) == 4
>>> # xdoctest: +REQUIRES(module:torch)
>>> subset = dets.tensor().take([2, 3, 5, 7])
>>> assert len(subset) == 4

__getitem__(self, index)

Fancy slicing / subset / indexing.

Note: scalar indices are always coerced into index lists of length 1.

Example

>>> import kwimage
>>> import kwarray
>>> dets = kwimage.Detections(boxes=kwimage.Boxes.random(10))
>>> indices = [2, 3, 5, 7]
>>> flags = kwarray.boolmask(indices, len(dets))
>>> assert dets[flags].data == dets[indices].data

property device(self)

If the backend is torch returns the data device, otherwise None

is_tensor(self)

is the backend fueled by torch?

is_numpy(self)

is the backend fueled by numpy?

numpy(self)

Converts tensors to numpy. Does not change memory if possible.

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> self = Detections.random(3).tensor()
>>> newself = self.numpy()
>>> self.scores[0] = 0
>>> assert newself.scores[0] == 0
>>> self.scores[0] = 1
>>> assert self.scores[0] == 1
>>> self.numpy().numpy()

property dtype(self)

tensor(self, device=ub.NoParam)

Converts numpy to tensors. Does not change memory if possible.

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> from kwimage.structs.detections import *
>>> self = Detections.random(3)
>>> newself = self.tensor()
>>> self.scores[0] = 0
>>> assert newself.scores[0] == 0
>>> self.scores[0] = 1
>>> assert self.scores[0] == 1
>>> self.tensor().tensor()

classmethod demo(Detections)

classmethod random(cls, num=10, scale=1.0, classes=3, keypoints=False, segmentations=False, tensor=False, rng=None)

Creates dummy data, suitable for use in tests and benchmarks

Parameters

• num (int) – number of boxes

• scale (float | tuple, default=1.0) – bounding image size

• classes (int | Sequence) – list of class labels or number of classes

• keypoints (bool, default=False) – if True include random keypoints for each box.

• segmentations (bool, default=False) – if True include random segmentations for each box.

• tensor (bool, default=False) – determines backend. DEPRECATED. Call tensor on resulting object instead.

• rng (np.random.RandomState) – random state

Example

>>> import kwimage
>>> dets = kwimage.Detections.random(keypoints='jagged')
>>> dets.data['keypoints'].data[0].data
>>> dets.data['keypoints'].meta
>>> dets = kwimage.Detections.random(keypoints='dense')
>>> dets = kwimage.Detections.random(keypoints='dense', segmentations=True).scale(1000)
>>> # xdoctest:+REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> dets.draw(setlim=True)

Example

>>> import kwimage
>>> dets = kwimage.Detections.random(
>>>     keypoints='jagged', segmentations=True, rng=0).scale(1000)
>>> print('dets = {}'.format(dets))
dets = <Detections(10)>
>>> dets.data['boxes'].quantize(inplace=True)
>>> print('dets.data = {}'.format(ub.repr2(
>>>     dets.data, nl=1, with_dtype=False, strvals=True)))
dets.data = {
    'boxes': <Boxes(xywh,
                 array([[548, 544,  55, 172],
                        [423, 645,  15, 247],
                        [791, 383, 173, 146],
                        [ 71,  87, 498, 839],
                        [ 20, 832, 759,  39],
                        [461, 780, 518,  20],
                        [118, 639,  26, 306],
                        [264, 414, 258, 361],
                        [ 18, 568, 439,  50],
                        [612, 616, 332,  66]], dtype=int32))>,
    'class_idxs': [1, 2, 0, 0, 2, 0, 0, 0, 0, 0],
    'keypoints': <PointsList(n=10)>,
    'scores': [0.3595079 , 0.43703195, 0.6976312 , 0.06022547, 0.66676672, 0.67063787,0.21038256, 0.1289263 , 0.31542835, 0.36371077],
    'segmentations': <SegmentationList(n=10)>,
}
>>> # xdoctest:+REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> dets.draw(setlim=True)

Example

>>> # Boxes position/shape within 0-1 space should be uniform.
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> fig = kwplot.figure(fnum=1, doclf=True)
>>> fig.gca().set_xlim(0, 128)
>>> fig.gca().set_ylim(0, 128)
>>> import kwimage
>>> kwimage.Detections.random(num=10, segmentations=True).scale(128).draw()

class kwimage.structs.Heatmap(data=None, meta=None, **kwargs)

Bases: kwimage.structs._generic.Spatial, _HeatmapDrawMixin, _HeatmapWarpMixin, _HeatmapAlgoMixin

Keeps track of a downscaled heatmap and how to transform it to overlay the original input image. Heatmaps generally are used to estimate class probabilites at each pixel. This data struction additionally contains logic to augment pixel with offset (dydx) and scale (diamter) information.

Variables

• data (Dict[str, ArrayLike]) –

  dictionary containing spatially aligned heatmap data. Valid keys are as follows.

  class_probs (ArrayLike[C, H, W] | ArrayLike[C, D, H, W]):

  A probability map for each class. C is the number of classes.

  offset (ArrayLike[2, H, W] | ArrayLike[3, D, H, W], optional):

  object center position offset in y,x / t,y,x coordinates

  diamter (ArrayLike[2, H, W] | ArrayLike[3, D, H, W], optional):

  object bounding box sizes in h,w / d,h,w coordinates

  keypoints (ArrayLike[2, K, H, W] | ArrayLike[3, K, D, H, W], optional):

  y/x offsets for K different keypoint classes

• meta (Dict[str, object]) –

  dictionary containing miscellanious metadata about the heatmap data. Valid keys are as follows.

  img_dims (Tuple[H, W] | Tuple[D, H, W]):

  original image dimension

  tf_data_to_image (skimage.transform._geometric.GeometricTransform):

  transformation matrix (typically similarity or affine) that projects the given, heatmap onto the image dimensions such that the image and heatmap are spatially aligned.

  classes (List[str] | ndsampler.CategoryTree):

  information about which index in data[‘class_probs’] corresponds to which semantic class.

• dims (Tuple) – dimensions of the heatmap (See `image_dims) for the original image dimensions.

• **kwargs – any key that is accepted by the data or meta dictionaries can be specified as a keyword argument to this class and it will be properly placed in the appropriate internal dictionary.

CommandLine:

xdoctest -m ~/code/kwimage/kwimage/structs/heatmap.py Heatmap –show

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> from kwimage.structs.heatmap import *  # NOQA
>>> import kwimage
>>> class_probs = kwimage.grab_test_image(dsize=(32, 32), space='gray')[None, ] / 255.0
>>> img_dims = (220, 220)
>>> tf_data_to_img = skimage.transform.AffineTransform(translation=(-18, -18), scale=(8, 8))
>>> self = Heatmap(class_probs=class_probs, img_dims=img_dims,
>>>                tf_data_to_img=tf_data_to_img)
>>> aligned = self.upscale()
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.imshow(aligned[0])
>>> kwplot.show_if_requested()

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> import kwimage
>>> self = Heatmap.random()
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> self.draw()

__datakeys__ = ['class_probs', 'offset', 'diameter', 'keypoints', 'class_idx', 'class_energy']

__metakeys__ = ['img_dims', 'tf_data_to_img', 'classes', 'kp_classes']

__spatialkeys__ = ['offset', 'diameter', 'keypoints']

__nice__(self)

__getitem__(self, index)

__len__(self)

property shape(self)

property bounds(self)

property dims(self)

space-time dimensions of this heatmap

is_numpy(self)

is_tensor(self)

property _impl(self)

Returns the internal tensor/numpy ArrayAPI implementation

Returns

kwarray.ArrayAPI

classmethod random(cls, dims=(10, 10), classes=3, diameter=True, offset=True, keypoints=False, img_dims=None, dets=None, nblips=10, noise=0.0, rng=None)

Creates dummy data, suitable for use in tests and benchmarks

Parameters

• dims (Tuple) – dimensions of the heatmap

• img_dims (Tuple) – dimensions of the image the heatmap corresponds to

Example

>>> from kwimage.structs.heatmap import *  # NOQA
>>> self = Heatmap.random((128, 128), img_dims=(200, 200),
>>>     classes=3, nblips=10, rng=0, noise=0.1)
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.imshow(self.colorize(0, imgspace=0), fnum=1, pnum=(1, 4, 1), doclf=1)
>>> kwplot.imshow(self.colorize(1, imgspace=0), fnum=1, pnum=(1, 4, 2))
>>> kwplot.imshow(self.colorize(2, imgspace=0), fnum=1, pnum=(1, 4, 3))
>>> kwplot.imshow(self.colorize(3, imgspace=0), fnum=1, pnum=(1, 4, 4))

Ignore:

self.detect(0).sort().non_max_supress()[-np.arange(1, 4)].draw() from kwimage.structs.heatmap import * # NOQA import xdev globals().update(xdev.get_func_kwargs(Heatmap.random))

Example

>>> # xdoctest: +REQUIRES(module:ndsampler)
>>> import kwimage
>>> self = kwimage.Heatmap.random(dims=(50, 200), dets='coco',
>>>                               keypoints=True)
>>> image = np.zeros(self.img_dims)
>>> # xdoctest: +REQUIRES(module:kwplot)
>>> toshow = self.draw_on(image, 1, vecs=True, kpts=0, with_alpha=0.85)
>>> # xdoctest: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.imshow(toshow)

Ignore:

>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.imshow(image)
>>> dets.draw()
>>> dets.data['keypoints'].draw(radius=6)
>>> dets.data['segmentations'].draw()

>>> self.draw()

property class_probs(self)

property offset(self)

property diameter(self)

property img_dims(self)

property tf_data_to_img(self)

property classes(self)

numpy(self)

Converts underlying data to numpy arrays

tensor(self, device=ub.NoParam)

Converts underlying data to torch tensors

kwimage.structs.smooth_prob(prob, k=3, inplace=False, eps=1e-09)

Smooths the probability map, but preserves the magnitude of the peaks.

Notes

even if inplace is true, we still need to make a copy of the input array, however, we do ensure that it is cleaned up before we leave the function scope.

sigma=0.8 @ k=3, sigma=1.1 @ k=5, sigma=1.4 @ k=7

class kwimage.structs.Mask(data=None, format=None)

Bases: ubelt.NiceRepr, _MaskConversionMixin, _MaskConstructorMixin, _MaskTransformMixin, _MaskDrawMixin

Manages a single segmentation mask and can convert to and from multiple formats including:

• bytes_rle - byte encoded run length encoding

• array_rle - raw run length encoding

• c_mask - c-style binary mask

• f_mask - fortran-style binary mask

Example

>>> # xdoc: +REQUIRES(--mask)
>>> # a ms-coco style compressed bytes rle segmentation
>>> segmentation = {'size': [5, 9], 'counts': ';?1B10O30O4'}
>>> mask = Mask(segmentation, 'bytes_rle')
>>> # convert to binary numpy representation
>>> binary_mask = mask.to_c_mask().data
>>> print(ub.repr2(binary_mask.tolist(), nl=1, nobr=1))
[0, 0, 0, 1, 1, 1, 1, 1, 0],
[0, 0, 1, 1, 1, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 1, 0],
[0, 0, 1, 1, 1, 0, 1, 1, 0],
[0, 0, 1, 1, 1, 0, 1, 1, 0],

property dtype(self)

__nice__(self)

classmethod random(Mask, rng=None, shape=(32, 32))

Create a random binary mask object

Parameters

• rng (int | RandomState | None) – the random seed

• shape (Tuple[int, int]) – the height / width of the returned mask

Returns

the random mask

Return type

Mask

Example

>>> import kwimage
>>> mask = kwimage.Mask.random()
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> mask.draw()
>>> kwplot.show_if_requested()

classmethod demo(cls)

Demo mask with holes and disjoint shapes

Returns

the demo mask

Return type

Mask

copy(self)

Performs a deep copy of the mask data

Returns

the copied mask

Return type

Mask

Example

>>> self = Mask.random(shape=(8, 8), rng=0)
>>> other = self.copy()
>>> assert other.data is not self.data

union(self, *others)

This can be used as a staticmethod or an instancemethod

Parameters

*others – multiple input masks to union

Returns

the unioned mask

Return type

Mask

Example

>>> # xdoc: +REQUIRES(--mask)
>>> from kwimage.structs.mask import *  # NOQA
>>> masks = [Mask.random(shape=(8, 8), rng=i) for i in range(2)]
>>> mask = Mask.union(*masks)
>>> print(mask.area)
>>> masks = [m.to_c_mask() for m in masks]
>>> mask = Mask.union(*masks)
>>> print(mask.area)

>>> masks = [m.to_bytes_rle() for m in masks]
>>> mask = Mask.union(*masks)
>>> print(mask.area)

Benchmark:

import ubelt as ub ti = ub.Timerit(100, bestof=10, verbose=2)

masks = [Mask.random(shape=(172, 172), rng=i) for i in range(2)]

for timer in ti.reset(‘native rle union’):

masks = [m.to_bytes_rle() for m in masks] with timer:

mask = Mask.union(*masks)

for timer in ti.reset(‘native cmask union’):

masks = [m.to_c_mask() for m in masks] with timer:

mask = Mask.union(*masks)

for timer in ti.reset(‘cmask->rle union’):

masks = [m.to_c_mask() for m in masks] with timer:

mask = Mask.union(*[m.to_bytes_rle() for m in masks])

intersection(self, *others)

This can be used as a staticmethod or an instancemethod

Parameters

*others – multiple input masks to intersect

Returns

the intersection of the masks

Return type

Mask

Example

>>> n = 3
>>> masks = [Mask.random(shape=(8, 8), rng=i) for i in range(n)]
>>> items = masks
>>> mask = Mask.intersection(*masks)
>>> areas = [item.area for item in items]
>>> print('areas = {!r}'.format(areas))
>>> print(mask.area)
>>> print(Mask.intersection(*masks).area / Mask.union(*masks).area)

property shape(self)

property area(self)

Returns the number of non-zero pixels

Returns

the number of non-zero pixels

Return type

int

Example

>>> self = Mask.demo()
>>> self.area
150

get_patch(self)

Extract the patch with non-zero data

Example

>>> # xdoc: +REQUIRES(--mask)
>>> from kwimage.structs.mask import *  # NOQA
>>> self = Mask.random(shape=(8, 8), rng=0)
>>> self.get_patch()

get_xywh(self)

Gets the bounding xywh box coordinates of this mask

Returns

x, y, w, h: Note we dont use a Boxes object because

a general singular version does not yet exist.

Return type

ndarray

Example

>>> # xdoc: +REQUIRES(--mask)
>>> self = Mask.random(shape=(8, 8), rng=0)
>>> self.get_xywh().tolist()
>>> self = Mask.random(rng=0).translate((10, 10))
>>> self.get_xywh().tolist()

Example

>>> # test empty case
>>> import kwimage
>>> self = kwimage.Mask(np.empty((0, 0), dtype=np.uint8), format='c_mask')
>>> assert self.get_xywh().tolist() == [0, 0, 0, 0]

Ignore:

>>> import kwimage
>>> self = kwimage.Mask(np.zeros((768, 768), dtype=np.uint8), format='c_mask')
>>> x_coords = np.array([621, 752])
>>> y_coords = np.array([366, 292])
>>> self.data[y_coords, x_coords] = 1
>>> self.get_xywh()

>>> # References:
>>> # https://stackoverflow.com/questions/33281957/faster-alternative-to-numpy-where
>>> # https://answers.opencv.org/question/4183/what-is-the-best-way-to-find-bounding-box-for-binary-mask/
>>> import timerit
>>> ti = timerit.Timerit(100, bestof=10, verbose=2)
>>> for timer in ti.reset('time'):
>>>     with timer:
>>>         y_coords, x_coords = np.where(self.data)
>>> #
>>> for timer in ti.reset('time'):
>>>     with timer:
>>>         cv2.findNonZero(data)

self.data = np.random.rand(800, 700) > 0.5

import timerit ti = timerit.Timerit(100, bestof=10, verbose=2) for timer in ti.reset(‘time’):

with timer:

y_coords, x_coords = np.where(self.data)

# for timer in ti.reset(‘time’):

with timer:

data = np.ascontiguousarray(self.data).astype(np.uint8) cv2_coords = cv2.findNonZero(data)

>>> poly = self.to_multi_polygon()

get_polygon(self)

DEPRECATED: USE to_multi_polygon

Returns a list of (x,y)-coordinate lists. The length of the list is equal to the number of disjoint regions in the mask.

Returns

polygon around each connected component of the

mask. Each ndarray is an Nx2 array of xy points.

Return type

List[ndarray]

Note

The returned polygon may not surround points that are only one pixel thick.

Example

>>> # xdoc: +REQUIRES(--mask)
>>> from kwimage.structs.mask import *  # NOQA
>>> self = Mask.random(shape=(8, 8), rng=0)
>>> polygons = self.get_polygon()
>>> print('polygons = ' + ub.repr2(polygons))
>>> polygons = self.get_polygon()
>>> self = self.to_bytes_rle()
>>> other = Mask.from_polygons(polygons, self.shape)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> image = np.ones(self.shape)
>>> image = self.draw_on(image, color='blue')
>>> image = other.draw_on(image, color='red')
>>> kwplot.imshow(image)

polygons = [

np.array([[6, 4],[7, 4]], dtype=np.int32), np.array([[0, 1],[0, 3],[2, 3],[2, 1]], dtype=np.int32),

]

to_mask(self, dims=None)

Converts to a mask object (which does nothing because this already is mask object!)

Returns

kwimage.Mask

to_boxes(self)

Returns the bounding box of the mask.

Returns

kwimage.Boxes

to_multi_polygon(self)

Returns a MultiPolygon object fit around this raster including disjoint pieces and holes.

Returns

vectorized representation

Return type

MultiPolygon

Example

>>> # xdoc: +REQUIRES(--mask)
>>> from kwimage.structs.mask import *  # NOQA
>>> self = Mask.demo()
>>> self = self.scale(5)
>>> multi_poly = self.to_multi_polygon()
>>> # xdoc: +REQUIRES(module:kwplot)
>>> # xdoc: +REQUIRES(--show)
>>> self.draw(color='red')
>>> multi_poly.scale(1.1).draw(color='blue')

>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> image = np.ones(self.shape)
>>> image = self.draw_on(image, color='blue')
>>> #image = other.draw_on(image, color='red')
>>> kwplot.imshow(image)
>>> multi_poly.draw()

Example

>>> import kwimage
>>> self = kwimage.Mask(np.empty((0, 0), dtype=np.uint8), format='c_mask')
>>> poly = self.to_multi_polygon()
>>> poly.to_multi_polygon()

Example

# Corner case, only two pixels are on >>> import kwimage >>> self = kwimage.Mask(np.zeros((768, 768), dtype=np.uint8), format=’c_mask’) >>> x_coords = np.array([621, 752]) >>> y_coords = np.array([366, 292]) >>> self.data[y_coords, x_coords] = 1 >>> poly = self.to_multi_polygon()

poly.to_mask(self.shape).data.sum()

self.to_array_rle().to_c_mask().data.sum() temp.to_c_mask().data.sum()

Example

>>> # TODO: how do we correctly handle the 1 or 2 point to a poly
>>> # case?
>>> import kwimage
>>> data = np.zeros((8, 8), dtype=np.uint8)
>>> data[0, 3:5] = 1
>>> data[7, 3:5] = 1
>>> data[3:5, 0:2] = 1
>>> self = kwimage.Mask.coerce(data)
>>> polys = self.to_multi_polygon()
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.imshow(data)
>>> polys.draw(border=True, linewidth=5, alpha=0.5, radius=0.2)

get_convex_hull(self)

Returns a list of xy points around the convex hull of this mask

Note

The returned polygon may not surround points that are only one pixel thick.

Example

>>> # xdoc: +REQUIRES(--mask)
>>> self = Mask.random(shape=(8, 8), rng=0)
>>> polygons = self.get_convex_hull()
>>> print('polygons = ' + ub.repr2(polygons))
>>> other = Mask.from_polygons(polygons, self.shape)

iou(self, other)

The area of intersection over the area of union

Todo

• [ ] Write plural Masks version of this class, which should

  be able to perform this operation more efficiently.

CommandLine:

xdoctest -m kwimage.structs.mask Mask.iou

Example

>>> # xdoc: +REQUIRES(--mask)
>>> self = Mask.demo()
>>> other = self.translate(1)
>>> iou = self.iou(other)
>>> print('iou = {:.4f}'.format(iou))
iou = 0.0830
>>> iou2 = self.intersection(other).area / self.union(other).area
>>> print('iou2 = {:.4f}'.format(iou2))

classmethod coerce(Mask, data, dims=None)

Attempts to auto-inspect the format of the data and conver to Mask

Parameters

• data – the data to coerce

• dims (Tuple) – required for certain formats like polygons height / width of the source image

Returns

the constructed mask object

Return type

Mask

Example

>>> # xdoc: +REQUIRES(--mask)
>>> segmentation = {'size': [5, 9], 'counts': ';?1B10O30O4'}
>>> polygon = [
>>>     [np.array([[3, 0],[2, 1],[2, 4],[4, 4],[4, 3],[7, 0]])],
>>>     [np.array([[2, 1],[2, 2],[4, 2],[4, 1]])],
>>> ]
>>> dims = (9, 5)
>>> mask = (np.random.rand(32, 32) > .5).astype(np.uint8)
>>> Mask.coerce(polygon, dims).to_bytes_rle()
>>> Mask.coerce(segmentation).to_bytes_rle()
>>> Mask.coerce(mask).to_bytes_rle()

_to_coco(self)

use to_coco instead

to_coco(self, style='orig')

Convert the Mask to a COCO json representation based on the current format.

A COCO mask is formatted as a run-length-encoding (RLE), of which there are two variants: (1) a array RLE, which is slightly more readable and extensible, and (2) a bytes RLE, which is slightly more concise. The returned format will depend on the current format of the Mask object. If it is in “bytes_rle” format, it will be returned in that format, otherwise it will be converted to the “array_rle” format and returned as such.

Parameters

style (str) – Does nothing for this particular method, exists for API compatibility and if alternate encoding styles are implemented in the future.

Returns

either a bytes-rle or array-rle encoding, depending

on the current mask format. The keys in this dictionary are as follows:

counts (List[int] | str): the array or bytes rle encoding

size (Tuple[int]): the height and width of the encoded mask

see note.

shape (Tuple[int]): only present in array-rle mode. This

is also the height/width of the underlying encoded array. This exists for semantic consistency with other kwimage conventions, and is not part of the original coco spec.

order (str): only present in array-rle mode.

Either C or F, indicating if counts is aranged in row-major or column-major order. For COCO-compatibility this is always returned in F (column-major) order.

binary (bool): only present in array-rle mode.

For COCO-compatibility this is always returned as False, indicating the mask only contains binary 0 or 1 values.

Return type

dict

Note

The output dictionary will contain a key named “size”, this is the only location in kwimage where “size” refers to a tuple in (height/width) order, in order to be backwards compatible with the original coco spec. In all other locations in kwimage a “size” will refer to a (width/height) ordered tuple.

SeeAlso:

func

kwimage.im_runlen.encode_run_length - backend function that does array-style run length encoding.

Example

>>> # xdoc: +REQUIRES(--mask)
>>> from kwimage.structs.mask import *  # NOQA
>>> self = Mask.demo()
>>> coco_data1 = self.toformat('array_rle').to_coco()
>>> coco_data2 = self.toformat('bytes_rle').to_coco()
>>> print('coco_data1 = {}'.format(ub.repr2(coco_data1, nl=1)))
>>> print('coco_data2 = {}'.format(ub.repr2(coco_data2, nl=1)))
coco_data1 = {
    'binary': True,
    'counts': [47, 5, 3, 1, 14, ... 1, 4, 19, 141],
    'order': 'F',
    'shape': (23, 32),
    'size': (23, 32),
}
coco_data2 = {
    'counts': '_153L;4EL...ON3060L0N060L0Nb0Y4',
    'size': [23, 32],
}

class kwimage.structs.MaskList

Bases: kwimage.structs._generic.ObjectList

Store and manipulate multiple masks, usually within the same image

to_polygon_list(self)

Converts all mask objects to multi-polygon objects

Returns

kwimage.PolygonList

to_segmentation_list(self)

Converts all items to segmentation objects

Returns

kwimage.SegmentationList

to_mask_list(self)

returns this object

Returns

kwimage.MaskList

class kwimage.structs.Points(data=None, meta=None, datakeys=None, metakeys=None, **kwargs)

Bases: kwimage.structs._generic.Spatial, _PointsWarpMixin

Stores multiple keypoints for a single object.

This stores both the geometry and the class metadata if available

Ignore:

meta = {

“names” = [‘head’, ‘nose’, ‘tail’], “skeleton” = [(0, 1), (0, 2)],

}

Example

>>> from kwimage.structs.points import *  # NOQA
>>> xy = np.random.rand(10, 2)
>>> pts = Points(xy=xy)
>>> print('pts = {!r}'.format(pts))

__datakeys__ = ['xy', 'class_idxs', 'visible']

__metakeys__ = ['classes']

__repr__

__nice__(self)

__len__(self)

property shape(self)

property xy(self)

classmethod random(Points, num=1, classes=None, rng=None)

Makes random points; typically for testing purposes

Example

>>> import kwimage
>>> self = kwimage.Points.random(classes=[1, 2, 3])
>>> self.data
>>> print('self.data = {!r}'.format(self.data))

is_numpy(self)

is_tensor(self)

_impl(self)

tensor(self, device=ub.NoParam)

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(10)
>>> self.tensor()

round(self, inplace=False)

Rounds data to the nearest integer

Parameters

inplace (bool, default=False) – if True, modifies this object

Example

>>> import kwimage
>>> self = kwimage.Points.random(3).scale(10)
>>> self.round()

numpy(self)

Example

>>> # xdoctest: +REQUIRES(module:torch)
>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(10)
>>> self.tensor().numpy().tensor().numpy()

draw_on(self, image, color='white', radius=None, copy=False)

CommandLine:

xdoctest -m ~/code/kwimage/kwimage/structs/points.py Points.draw_on –show

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.points import *  # NOQA
>>> s = 128
>>> image = np.zeros((s, s))
>>> self = Points.random(10).scale(s)
>>> image = self.draw_on(image)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.autompl()
>>> kwplot.imshow(image)
>>> self.draw(radius=3, alpha=.5)
>>> kwplot.show_if_requested()

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.points import *  # NOQA
>>> s = 128
>>> image = np.zeros((s, s))
>>> self = Points.random(10).scale(s)
>>> image = self.draw_on(image, radius=3, color='distinct')
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.autompl()
>>> kwplot.imshow(image)
>>> self.draw(radius=3, alpha=.5, color='classes')
>>> kwplot.show_if_requested()

Example

>>> import kwimage
>>> s = 32
>>> self = kwimage.Points.random(10).scale(s)
>>> color = 'blue'
>>> # Test drawong on all channel + dtype combinations
>>> im3 = np.zeros((s, s, 3), dtype=np.float32)
>>> im_chans = {
>>>     'im3': im3,
>>>     'im1': kwimage.convert_colorspace(im3, 'rgb', 'gray'),
>>>     'im4': kwimage.convert_colorspace(im3, 'rgb', 'rgba'),
>>> }
>>> inputs = {}
>>> for k, im in im_chans.items():
>>>     inputs[k + '_01'] = (kwimage.ensure_float01(im.copy()), {'radius': None})
>>>     inputs[k + '_255'] = (kwimage.ensure_uint255(im.copy()), {'radius': None})
>>> outputs = {}
>>> for k, v in inputs.items():
>>>     im, kw = v
>>>     outputs[k] = self.draw_on(im, color=color, **kw)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.figure(fnum=2, doclf=True)
>>> kwplot.autompl()
>>> pnum_ = kwplot.PlotNums(nCols=2, nRows=len(inputs))
>>> for k in inputs.keys():
>>>     kwplot.imshow(inputs[k][0], fnum=2, pnum=pnum_(), title=k)
>>>     kwplot.imshow(outputs[k], fnum=2, pnum=pnum_(), title=k)
>>> kwplot.show_if_requested()

draw(self, color='blue', ax=None, alpha=None, radius=1, **kwargs)

TODO: can use kwplot.draw_points

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.points import *  # NOQA
>>> pts = Points.random(10)
>>> # xdoc: +REQUIRES(--show)
>>> pts.draw(radius=0.01)

>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(10, classes=['a', 'b', 'c'])
>>> self.draw(radius=0.01, color='classes')

compress(self, flags, axis=0, inplace=False)

Filters items based on a boolean criterion

Example

>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(4)
>>> flags = [1, 0, 1, 1]
>>> other = self.compress(flags)
>>> assert len(self) == 4
>>> assert len(other) == 3

>>> # xdoctest: +REQUIRES(module:torch)
>>> other = self.tensor().compress(flags)
>>> assert len(other) == 3

take(self, indices, axis=0, inplace=False)

Takes a subset of items at specific indices

Example

>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(4)
>>> indices = [1, 3]
>>> other = self.take(indices)
>>> assert len(self) == 4
>>> assert len(other) == 2

>>> # xdoctest: +REQUIRES(module:torch)
>>> other = self.tensor().take(indices)
>>> assert len(other) == 2

classmethod concatenate(cls, points, axis=0)

to_coco(self, style='orig')

Converts to an mscoco-like representation

Note

items that are usually id-references to other objects may need to be rectified.

Parameters

style (str) – either orig, new, new-id, or new-name

Returns

mscoco-like representation

Return type

Dict

Example

>>> from kwimage.structs.points import *  # NOQA
>>> self = Points.random(4, classes=['a', 'b'])
>>> orig = self._to_coco(style='orig')
>>> print('orig = {!r}'.format(orig))
>>> new_name = self._to_coco(style='new-name')
>>> print('new_name = {}'.format(ub.repr2(new_name, nl=-1)))
>>> # xdoctest: +REQUIRES(module:ndsampler)
>>> import ndsampler
>>> self.meta['classes'] = ndsampler.CategoryTree.coerce(self.meta['classes'])
>>> new_id = self._to_coco(style='new-id')
>>> print('new_id = {}'.format(ub.repr2(new_id, nl=-1)))

_to_coco(self, style='orig')

See to_coco

classmethod coerce(cls, data)

Attempt to coerce data into a Points object

classmethod _from_coco(cls, coco_kpts, class_idxs=None, classes=None)

classmethod from_coco(cls, coco_kpts, class_idxs=None, classes=None, warn=False)

Parameters

• coco_kpts (list | dict) – either the original list keypoint encoding or the new dict keypoint encoding.

• class_idxs (list) – only needed if using old style

• classes (list | CategoryTree) – list of all keypoint category names

• warn (bool, default=False) – if True raise warnings

Example

>>> ##
>>> classes = ['mouth', 'left-hand', 'right-hand']
>>> coco_kpts = [
>>>     {'xy': (0, 0), 'visible': 2, 'keypoint_category': 'left-hand'},
>>>     {'xy': (1, 2), 'visible': 2, 'keypoint_category': 'mouth'},
>>> ]
>>> Points.from_coco(coco_kpts, classes=classes)
>>> # Test without classes
>>> Points.from_coco(coco_kpts)
>>> # Test without any category info
>>> coco_kpts2 = [ub.dict_diff(d, {'keypoint_category'}) for d in coco_kpts]
>>> Points.from_coco(coco_kpts2)
>>> # Test without category instead of keypoint_category
>>> coco_kpts3 = [ub.map_keys(lambda x: x.replace('keypoint_', ''), d) for d in coco_kpts]
>>> Points.from_coco(coco_kpts3)
>>> #
>>> # Old style
>>> coco_kpts = [0, 0, 2, 0, 1, 2]
>>> Points.from_coco(coco_kpts)
>>> # Fail case
>>> coco_kpts4 = [{'xy': [4686.5, 1341.5], 'category': 'dot'}]
>>> Points.from_coco(coco_kpts4, classes=[])

Example

>>> # xdoctest: +REQUIRES(module:ndsampler)
>>> import ndsampler
>>> classes = ndsampler.CategoryTree.from_coco([
>>>     {'name': 'mouth', 'id': 2}, {'name': 'left-hand', 'id': 3}, {'name': 'right-hand', 'id': 5}
>>> ])
>>> coco_kpts = [
>>>     {'xy': (0, 0), 'visible': 2, 'keypoint_category_id': 5},
>>>     {'xy': (1, 2), 'visible': 2, 'keypoint_category_id': 2},
>>> ]
>>> pts = Points.from_coco(coco_kpts, classes=classes)
>>> assert pts.data['class_idxs'].tolist() == [2, 0]

class kwimage.structs.PointsList

Bases: kwimage.structs._generic.ObjectList

Stores a list of Points, each item usually corresponds to a different object.

Notes

# TODO: when the data is homogenous we can use a more efficient # representation, otherwise we have to use heterogenous storage.

class kwimage.structs.MultiPolygon

Bases: kwimage.structs._generic.ObjectList

Data structure for storing multiple polygons (typically related to the same underlying but potentitally disjoing object)

Variables

data (List[Polygon]) –

classmethod random(self, n=3, n_holes=0, rng=None, tight=False)

Create a random MultiPolygon

Returns

MultiPolygon

fill(self, image, value=1)

Inplace fill in an image based on this multi-polyon.

Parameters

• image (ndarray) – image to draw on (inplace)

• value (int | Tuple[int], default=1) – value fill in with

Returns

the image that has been modified in place

Return type

ndarray

to_multi_polygon(self)

to_boxes(self)

Deprecated: lossy conversion use ‘bounding_box’ instead

bounding_box(self)

Return the bounding box of the multi polygon

Returns

a Boxes object with one box that encloses all

polygons

Return type

kwimage.Boxes

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> self = MultiPolygon.random(rng=0, n=10)
>>> boxes = self.to_boxes()
>>> sub_boxes = [d.to_boxes() for d in self.data]
>>> areas1 = np.array([s.intersection(boxes).area[0] for s in sub_boxes])
>>> areas2 = np.array([s.area[0] for s in sub_boxes])
>>> assert np.allclose(areas1, areas2)

to_mask(self, dims=None)

Returns a mask object indication regions occupied by this multipolygon

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> s = 100
>>> self = MultiPolygon.random(rng=0).scale(s)
>>> dims = (s, s)
>>> mask = self.to_mask(dims)

>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> from matplotlib import pyplot as pl
>>> ax = plt.gca()
>>> ax.set_xlim(0, s)
>>> ax.set_ylim(0, s)
>>> self.draw(color='red', alpha=.4)
>>> mask.draw(color='blue', alpha=.4)

to_relative_mask(self)

Returns a translated mask such the mask dimensions are minimal.

In other words, we move the polygon all the way to the top-left and return a mask just big enough to fit the polygon.

Returns

Mask

classmethod coerce(cls, data, dims=None)

Attempts to construct a MultiPolygon instance from the input data

See Mask.coerce

to_shapely(self)

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> # xdoc: +REQUIRES(module:shapely)
>>> from kwimage.structs.polygon import *  # NOQA
>>> self = MultiPolygon.random(rng=0)
>>> geom = self.to_shapely()
>>> print('geom = {!r}'.format(geom))

classmethod from_shapely(MultiPolygon, geom)

Convert a shapely polygon or multipolygon to a kwimage.MultiPolygon

classmethod from_geojson(MultiPolygon, data_geojson)

Convert a geojson polygon or multipolygon to a kwimage.MultiPolygon

Example

>>> import kwimage
>>> orig = kwimage.MultiPolygon.random()
>>> data_geojson = orig.to_geojson()
>>> self = kwimage.MultiPolygon.from_geojson(data_geojson)

to_geojson(self)

Converts polygon to a geojson structure

classmethod from_coco(cls, data, dims=None)

Accepts either new-style or old-style coco multi-polygons

_to_coco(self, style='orig')

to_coco(self, style='orig')

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> self = MultiPolygon.random(1, rng=0)
>>> self.to_coco()

swap_axes(self, inplace=False)

class kwimage.structs.Polygon(data=None, meta=None, datakeys=None, metakeys=None, **kwargs)

Bases: kwimage.structs._generic.Spatial, _PolyArrayBackend, _PolyWarpMixin, ubelt.NiceRepr

Represents a single polygon as set of exterior boundary points and a list of internal polygons representing holes.

By convention exterior boundaries should be counterclockwise and interior holes should be clockwise.

Example

>>> import kwimage
>>> data = {
>>>     'exterior': np.array([[13,  1], [13, 19], [25, 19], [25,  1]]),
>>>     'interiors': [
>>>         np.array([[13, 13], [14, 12], [24, 12], [25, 13], [25, 18],
>>>                   [24, 19], [14, 19], [13, 18]]),
>>>         np.array([[13,  2], [14,  1], [24,  1], [25, 2], [25, 11],
>>>                   [24, 12], [14, 12], [13, 11]])]
>>> }
>>> self = kwimage.Polygon(**data)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> self.draw(setlim=True)

Example

>>> import kwimage
>>> self = kwimage.Polygon.random(
>>>     n=5, n_holes=1, convex=False, rng=0)
>>> print('self = {}'.format(self))
self = <Polygon({
    'exterior': <Coords(data=
                    array([[0.30371392, 0.97195856],
                           [0.24372304, 0.60568445],
                           [0.21408694, 0.34884262],
                           [0.5799477 , 0.44020379],
                           [0.83720288, 0.78367234]]))>,
    'interiors': [<Coords(data=
                     array([[0.50164209, 0.83520279],
                            [0.25835064, 0.40313428],
                            [0.28778562, 0.74758761],
                            [0.30341266, 0.93748088]]))>],
})>
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> self.draw(setlim=True)

__datakeys__ = ['exterior', 'interiors']

__metakeys__ = ['classes']

property exterior(self)

property interiors(self)

__nice__(self)

classmethod circle(cls, xy, r, resolution=64)

Create a circular polygon

Example

>>> xy = (0.5, 0.5)
>>> r = .3
>>> poly = Polygon.circle(xy, r)

classmethod random(cls, n=6, n_holes=0, convex=True, tight=False, rng=None)

Parameters

• n (int) – number of points in the polygon (must be 3 or more)

• n_holes (int) – number of holes

• tight (bool, default=False) – fits the minimum and maximum points between 0 and 1

• convex (bool, default=True) – force resulting polygon will be convex (may remove exterior points)

CommandLine:

xdoctest -m kwimage.structs.polygon Polygon.random

Example

>>> rng = None
>>> n = 4
>>> n_holes = 1
>>> cls = Polygon
>>> self = Polygon.random(n=n, rng=rng, n_holes=n_holes, convex=1)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.figure(fnum=1, doclf=True)
>>> kwplot.autompl()
>>> self.draw()

References

https://gis.stackexchange.com/questions/207731/random-multipolygon https://stackoverflow.com/questions/8997099/random-polygon https://stackoverflow.com/questions/27548363/from-voronoi-tessellation-to-shapely-polygons https://stackoverflow.com/questions/8997099/algorithm-to-generate-random-2d-polygon

_impl(self)

to_mask(self, dims=None)

Convert this polygon to a mask

Todo

• [ ] currently not efficient

Parameters

dims (Tuple) – height and width of the output mask

Returns

kwimage.Mask

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random(n_holes=1).scale(128)
>>> mask = self.to_mask((128, 128))
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> mask.draw(color='blue')
>>> mask.to_multi_polygon().draw(color='red', alpha=.5)

to_relative_mask(self)

Returns a translated mask such the mask dimensions are minimal.

In other words, we move the polygon all the way to the top-left and return a mask just big enough to fit the polygon.

Returns

kwimage.Mask

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random().scale(8).translate(100, 100)
>>> mask = self.to_relative_mask()
>>> assert mask.shape <= (8, 8)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> mask.draw(color='blue')
>>> mask.to_multi_polygon().draw(color='red', alpha=.5)

fill(self, image, value=1)

Inplace fill in an image based on this polyon.

Parameters

• image (ndarray) – image to draw on

• value (int | Tuple[int], default=1) – value fill in with

Returns

the image that has been modified in place

Return type

ndarray

_to_cv_countours(self)

OpenCV polygon representation, which is a list of points. Holes are implicitly represented. When another polygon is drawn over an existing polyon via cv2.fillPoly

Returns

where each ndarray is of shape [N, 1, 2],

where N is the number of points on the boundary, the middle dimension is always 1, and the trailing dimension represents x and y coordinates respectively.

Return type

List[ndarray]

classmethod coerce(Polygon, data)

Try to autodetermine format of input polygon and coerce it into a kwimage.Polygon.

Parameters

data (object) – some type of data that can be interpreted as a polygon.

Returns

kwimage.Polygon

Example

>>> import kwimage
>>> self = kwimage.Polygon.random()
>>> self.coerce(self)
>>> self.coerce(self.exterior)
>>> self.coerce(self.exterior.data)
>>> self.coerce(self.data)
>>> self.coerce(self.to_geojson())

classmethod from_shapely(Polygon, geom)

Convert a shapely polygon to a kwimage.Polygon

Parameters

geom (shapely.geometry.polygon.Polygon) – a shapely polygon

Returns

kwimage.Polygon

classmethod from_wkt(Polygon, data)

Convert a WKT string to a kwimage.Polygon

Parameters

data (str) – a WKT polygon string

Returns

kwimage.Polygon

Example

>>> import kwimage
>>> data = 'POLYGON ((0.11 0.61, 0.07 0.588, 0.015 0.50, 0.11 0.61))'
>>> self = kwimage.Polygon.from_wkt(data)
>>> assert len(self.exterior) == 4

classmethod from_geojson(Polygon, data_geojson)

Convert a geojson polygon to a kwimage.Polygon

Parameters

data_geojson (dict) – geojson data

References

https://geojson.org/geojson-spec.html

Example

>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random(n_holes=2)
>>> data_geojson = self.to_geojson()
>>> new = Polygon.from_geojson(data_geojson)

to_shapely(self)

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> # xdoc: +REQUIRES(module:shapely)
>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random(n_holes=1)
>>> self = self.scale(100)
>>> geom = self.to_shapely()
>>> print('geom = {!r}'.format(geom))

to_geojson(self)

Converts polygon to a geojson structure

Returns

Dict[str, object]

Example

>>> import kwimage
>>> self = kwimage.Polygon.random()
>>> print(self.to_geojson())

to_wkt(self)

Convert a kwimage.Polygon to WKT string

Example

>>> import kwimage
>>> self = kwimage.Polygon.random()
>>> print(self.to_wkt())

classmethod from_coco(cls, data, dims=None)

Accepts either new-style or old-style coco polygons

_to_coco(self, style='orig')

to_coco(self, style='orig')

Returns

coco-style polygons

Return type

List | Dict

to_multi_polygon(self)

to_boxes(self)

Deprecated: lossy conversion use ‘bounding_box’ instead

property centroid(self)

bounding_box(self)

Returns an axis-aligned bounding box for the segmentation

Returns

kwimage.Boxes

bounding_box_polygon(self)

Returns an axis-aligned bounding polygon for the segmentation.

Notes

This Polygon will be a Box, not a convex hull! Use shapely for convex hulls.

Returns

kwimage.Polygon

copy(self)

clip(self, x_min, y_min, x_max, y_max, inplace=False)

Clip polygon to image boundaries.

Example

>>> from kwimage.structs.polygon import *
>>> self = Polygon.random().scale(10).translate(-1)
>>> self2 = self.clip(1, 1, 3, 3)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> self2.draw(setlim=True)

draw_on(self, image, color='blue', fill=True, border=False, alpha=1.0, copy=False)

Rasterizes a polygon on an image. See draw for a vectorized matplotlib version.

Parameters

• image (ndarray) – image to raster polygon on.

• color (str | tuple) – data coercable to a color

• fill (bool, default=True) – draw the center mass of the polygon

• border (bool, default=False) – draw the border of the polygon

• alpha (float, default=1.0) – polygon transparency (setting alpha < 1 makes this function much slower).

• copy (bool, default=False) – if False only copies if necessary

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random(n_holes=1).scale(128)
>>> image = np.zeros((128, 128), dtype=np.float32)
>>> image = self.draw_on(image)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.imshow(image, fnum=1)

Example

>>> import kwimage
>>> color = 'blue'
>>> self = kwimage.Polygon.random(n_holes=1).scale(128)
>>> image = np.zeros((128, 128), dtype=np.float32)
>>> # Test drawong on all channel + dtype combinations
>>> im3 = np.random.rand(128, 128, 3)
>>> im_chans = {
>>>     'im3': im3,
>>>     'im1': kwimage.convert_colorspace(im3, 'rgb', 'gray'),
>>>     'im4': kwimage.convert_colorspace(im3, 'rgb', 'rgba'),
>>> }
>>> inputs = {}
>>> for k, im in im_chans.items():
>>>     inputs[k + '_01'] = (kwimage.ensure_float01(im.copy()), {'alpha': None})
>>>     inputs[k + '_255'] = (kwimage.ensure_uint255(im.copy()), {'alpha': None})
>>>     inputs[k + '_01_a'] = (kwimage.ensure_float01(im.copy()), {'alpha': 0.5})
>>>     inputs[k + '_255_a'] = (kwimage.ensure_uint255(im.copy()), {'alpha': 0.5})
>>> outputs = {}
>>> for k, v in inputs.items():
>>>     im, kw = v
>>>     outputs[k] = self.draw_on(im, color=color, **kw)
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.figure(fnum=2, doclf=True)
>>> kwplot.autompl()
>>> pnum_ = kwplot.PlotNums(nCols=2, nRows=len(inputs))
>>> for k in inputs.keys():
>>>     kwplot.imshow(inputs[k][0], fnum=2, pnum=pnum_(), title=k)
>>>     kwplot.imshow(outputs[k], fnum=2, pnum=pnum_(), title=k)
>>> kwplot.show_if_requested()

draw(self, color='blue', ax=None, alpha=1.0, radius=1, setlim=False, border=False, linewidth=2)

Draws polygon in a matplotlib axes. See draw_on for in-memory image modification.

Parameters

• setlim (bool) – if True ensures the limits of the axes contains the polygon

• color (str | Tuple) – coercable color

• alpha (float) – fill transparency

• setlim (bool) – if True, modify the x and y limits of the matplotlib axes such that the polygon is can be seen.

• border (bool, default=False) – if True, draws an edge border on the polygon.

• linewidth (bool) – width of the border

Todo

• [ ] Rework arguments in favor of matplotlib standards

Example

>>> # xdoc: +REQUIRES(module:kwplot)
>>> from kwimage.structs.polygon import *  # NOQA
>>> self = Polygon.random(n_holes=1)
>>> self = self.scale(100)
>>> # xdoc: +REQUIRES(--show)
>>> self.draw()
>>> import kwplot
>>> kwplot.autompl()
>>> from matplotlib import pyplot as plt
>>> kwplot.figure(fnum=2)
>>> self.draw(setlim=True)

_ensure_vertex_order(self, inplace=False)

Fixes vertex ordering so the exterior ring is CCW and the interior rings are CW.

Example

>>> import kwimage
>>> self = kwimage.Polygon.random(n=3, n_holes=2, rng=0)
>>> print('self = {!r}'.format(self))
>>> new = self._ensure_vertex_order()
>>> print('new = {!r}'.format(new))

>>> self = kwimage.Polygon.random(n=3, n_holes=2, rng=0).swap_axes()
>>> print('self = {!r}'.format(self))
>>> new = self._ensure_vertex_order()
>>> print('new = {!r}'.format(new))

class kwimage.structs.PolygonList

Bases: kwimage.structs._generic.ObjectList

Stores and allows manipluation of multiple polygons, usually within the same image.

to_mask_list(self, dims=None)

Converts all items to masks

to_polygon_list(self)

to_segmentation_list(self)

Converts all items to segmentation objects

swap_axes(self, inplace=False)

to_geojson(self, as_collection=False)

Converts a list of polygons/multipolygons to a geojson structure

Parameters

as_collection (bool) – if True, wraps the polygon geojson items in a geojson feature collection, otherwise just return a list of items.

Returns

items or geojson data

Return type

List[Dict] | Dict

Example

>>> import kwimage
>>> data = [kwimage.Polygon.random(),
>>>         kwimage.Polygon.random(n_holes=1),
>>>         kwimage.MultiPolygon.random(n_holes=1),
>>>         kwimage.MultiPolygon.random()]
>>> self = kwimage.PolygonList(data)
>>> geojson = self.to_geojson(as_collection=True)
>>> items = self.to_geojson(as_collection=False)
>>> print('geojson = {}'.format(ub.repr2(geojson, nl=-2, precision=1)))
>>> print('items = {}'.format(ub.repr2(items, nl=-2, precision=1)))

class kwimage.structs.Segmentation(data, format=None)

Bases: _WrapperObject

Either holds a MultiPolygon, Polygon, or Mask

Parameters

• data (object) – the underlying object

• format (str) – either ‘mask’, ‘polygon’, or ‘multipolygon’

classmethod random(cls, rng=None)

Example

>>> self = Segmentation.random()
>>> print('self = {!r}'.format(self))
>>> # xdoc: +REQUIRES(--show)
>>> import kwplot
>>> kwplot.autompl()
>>> kwplot.figure(fnum=1, doclf=True)
>>> self.draw()
>>> kwplot.show_if_requested()

to_multi_polygon(self)

to_mask(self, dims=None)

property meta(self)

classmethod coerce(cls, data, dims=None)

class kwimage.structs.SegmentationList

Bases: kwimage.structs._generic.ObjectList

Store and manipulate multiple segmentations (masks or polygons), usually within the same image

to_polygon_list(self)

Converts all mask objects to multi-polygon objects

to_mask_list(self, dims=None)

Converts all mask objects to multi-polygon objects

to_segmentation_list(self)

classmethod coerce(cls, data)

Interpret data as a list of Segmentations