kwimage.structs.segmentation module

Generic segmentation object that can use either a Mask or (Multi)Polygon backend.

class kwimage.structs.segmentation._WrapperObject

Bases: NiceRepr

draw(*args, **kw)

draw_on(*args, **kw)

See help(self.data.draw_on)

warp(*args, **kw)

translate(*args, **kw)

scale(*args, **kw)

to_coco(*args, **kw)

numpy(*args, **kw)

tensor(*args, **kw)

class kwimage.structs.segmentation.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(rng=None)

Example

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

to_multi_polygon()

to_mask(dims=None, pixels_are='points')

property meta

classmethod coerce(data, dims=None)

class kwimage.structs.segmentation.SegmentationList(data, meta=None)

Bases: ObjectList

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

to_polygon_list()

Converts all mask objects to multi-polygon objects

to_mask_list(dims=None, pixels_are='points')

Converts all mask objects to multi-polygon objects

to_segmentation_list()

classmethod coerce(data)

Interpret data as a list of Segmentations

kwimage.structs.segmentation._coerce_coco_segmentation(data, dims=None)

Attempts to auto-inspect the format of segmentation data

Parameters:

• data – the data to coerce

  2D-C-ndarray -> C_MASK 2D-F-ndarray -> F_MASK

  Dict(counts=bytes) -> BYTES_RLE Dict(counts=ndarray) -> ARRAY_RLE

  Dict(exterior=ndarray) -> ARRAY_RLE

  # List[List[int]] -> Polygon List[int] -> Polygon List[Dict] -> MultPolygon

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

Todo

• [ ] Handle WKT

Returns:

Mask | Polygon | MultiPolygon | Segmentation - depending on which is appropriate

Example

>>> segmentation = {'size': [5, 9], 'counts': ';?1B10O30O4'}
>>> dims = (9, 5)
>>> raw_mask = (np.random.rand(32, 32) > .5).astype(np.uint8)
>>> _coerce_coco_segmentation(segmentation)
>>> _coerce_coco_segmentation(raw_mask)

>>> coco_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]]),
>>> ]
>>> self = _coerce_coco_segmentation(coco_polygon, dims)
>>> print('self = {!r}'.format(self))
>>> coco_polygon = [
>>>     np.array([[3, 0],[2, 1],[2, 4],[4, 4],[4, 3],[7, 0]]),
>>> ]
>>> self = _coerce_coco_segmentation(coco_polygon, dims)
>>> print('self = {!r}'.format(self))