# -*- coding: utf-8 -*-
"""This module implements a Python representation of the CropObject,
the basic unit of annotation. See the :class:`CropObject` documentation."""
from __future__ import print_function, unicode_literals, division
from builtins import zip
from builtins import map
from builtins import str
from builtins import range
from builtins import object
import copy
import itertools
import logging
from typing import Any, Optional, List, Union, Tuple
import numpy
from muscima.utils import compute_connected_components
__version__ = "1.0"
__author__ = "Jan Hajic jr."
CROPOBJECT_MASK_ORDER = 'C'
#: The CropObject mask uses this numpy ordering when flattening the data.
##############################################################################
[docs]class CropObject(object):
"""One annotated object.
The CropObject represents one instance of an annotation. It implements
the following attributes:
* ``objid``: the unique number of the given annotation instance in the set
of annotations encoded in the containing `CropObjectList`.
* ``uid``: the global unique identifier of the annotation instance. String.
See :meth:`CropObject.parse_uid` method for format details.
* ``clsname``: the name of the label that was given to the annotation
(this is the human-readable string such as ``notehead-full``).
* ``top``: the vertical dimension (row) of the upper left corner pixel.
* ``left``: the horizontal dimension (column) of the upper left corner pixel.
* ``bottom``: the vertical dimension (row) of the lower right corner pixel + 1,
so that you can index the corresponding image rows using
``img[c.top:c.bottom]``.
* ``right``: the horizontal dimension (row) of the lower right corner pixel + 1,
so that you can index the corresponding image columns using
``img[:, c.left:c.right]``.
* ``width``: the amount of rows that the CropObject spans.
* ``height``: the amount of columns that the CropObject spans.
* ``mask``: a binary (0/1) numpy array that denotes the area within the
CropObject's bounding box (specified by ``top``, ``left``, ``height``
and ``width``) that the CropObject actually occupies. If the mask is
``None``, the object is understood to occupy the entire bounding box.
* ``data``: a dictionary that can be empty, or can contain anything. It is
generated from the optional ``<Data>`` element of a CropObject.
Constructing a simple CropObject that consists of the "b"-like flat music
notation symbol (never mind the ``uid`` for now):
>>> top = 10
>>> left = 15
>>> height = 10
>>> width = 4
>>> mask = numpy.array([[1, 1, 0, 0],
... [1, 0, 0, 0],
... [1, 0, 0, 0],
... [1, 0, 0, 0],
... [1, 0, 1, 1],
... [1, 1, 1, 1],
... [1, 0, 0, 1],
... [1, 0, 1, 1],
... [1, 1, 1, 0],
... [0, 1, 0, 0]])
>>> clsname = 'flat'
>>> uid = 'MUSCIMA++_1.0___muscima.cropobject.CropObject.doctest___0'
>>> c = CropObject(objid=0, clsname=clsname,
... top=top, left=left, height=height, width=width,
... inlinks=[], outlinks=[],
... mask=mask,
... uid=uid)
CropObjects can also form graphs, using the following attributes:
* ``outlinks``: Outgoing edges. A list of integers; it is assumed they are
valid ``objid`` within the same global/doc namespace.
* ``inlinks``: Incoming edges. A list of integers; it is assumed they are
valid ``objid`` within the same global/doc namespace.
So far, CropObject graphs do not support multiple relationship types.
**Unique identification**
The ``uid`` serves to identify the CropObject uniquely,
at least within the MUSCIMA dataset system. (We anticipate further
versions of the dataset, and need to plan for that.)
To uniquely identify a CropObject, we need three "levels":
* The "global", **dataset-level identification**: which dataset is this
CropObject coming from? (For this dataset: ``MUSCIMA++_1.0``)
* The "local", **document-level identification**: which document
(within the given dataset) is this CropObject coming from?
For MUSCIMA++ 1.0, this will usually be a string like
``CVC-MUSCIMA_W-35_N-08_D-ideal``, derived from the filename
under which the CropObjectList containing the given CropObject
is stored.
* The **within-document identification**, which is identical
to the ``objid``.
These three components are joined together into one string by
a delimiter: ``___``
The full ``uid`` of a CropObject then might look like this::
MUSCIMA-pp_1.0___CVC-MUSCIMA_W-35_N-08_D-ideal___611
You will need to use UIDs whenever you are combining CropObjects
from different documents, and/or datasets. (If you are really combining
datasets, make sure you know what you are doing -- some annotation
instructions may change between versions, so objects of the same class
might not exactly correspond to each other...) The dataset and document
names are available through appropriate instance attributes:
>>> c.doc
'muscima.cropobject.CropObject.doctest'
>>> c.dataset
'MUSCIMA++_1.0'
If you supply no ``uid`` at initialization time, a default UID will
be used:
>>> c.default_uid
'MUSCIMA_DEFAULT_DATASET_PLACEHOLDER___default-document___0'
(Don't abuse the default, though! It's intended just for transitioning
documents without UIDs to those that have them.)
On the other hand, the ``objid`` is a field intended to uniquely identify
a CropObject within the scope of one CropObject list (one annotation
document).
.. caution::
The scope of unique identification within MUSCIMA++ is only within
a ``<CropObjectList>``. Don't use ``objid`` to mix CropObjects from
multiple files!
**CropObjects and images**
CropObjects and images are not tightly bound. This is because the same
object can apply to multiple images: in the case of the CVC-MUSCIMA dataset,
for example, the same CropObjects are present both in the full image
and in the staff-less image. The limitation here is that CropObjects
are based on exact pixels, so in order to retain validity, the images
must correspond to each other exactly, as "layers".
Because CropObjects do not correspond to any given image, there is
no facility in the data format to link them to a specific one. You have to
take care of matching CropObject annotations to the right images by yourself.
The ``CropObject`` class implements some interactions with images.
To recover the area corresponding to a CropObject `c`, use:
>>> if c.mask is not None: crop = img[c.top:c.bottom, c.left:c.right] * c.mask #doctest: +SKIP
>>> if c.mask is None: crop = img[c.top:c.bottom, c.left:c.right] #doctest: +SKIP
Because this is clunky, we have implemented the following to get the crop:
>>> crop = c.project_to(img) #doctest: +SKIP
And to get the CropObject projected onto the entire image:
>>> crop = c.project_on(img) #doctest: +SKIP
Above, note the multiplicative role of the mask: while we typically would
expect the mask to be binary, in principle, this is not strictly necessary.
You could supply a different mask interpration, such as probabilistic.
However, we strongly advise not to misuse this feature unless you have
a really good reason; remember that the CropObject is supposed to represent
an annotation of a given image. (One possible use for a non-binary mask
that we can envision is aggregating multiple annotations of the same
image.)
For visualization, there is a more sophisticated method that renders
the CropObject as a transparent colored transparent rectangle over
an RGB image. (NOTE: this really changes the input image!)
>>> c_obj.render(img) #doctest: +SKIP
>>> plt.imshow(img); plt.show() #doctest: +SKIP
However, `CropObject.render()` currently does not support rendering
the mask.
**Disambiguating class names**
Since the class names are present
through the ``clsname`` attribute (``<MLClassName>`` element),
matching the list is no longer necessary for general understanding
of the file. The MLClassList file serves as a disambiguation tool:
there may be multiple annotation projects that use the same names
but maybe define them differently and use different guidelines,
and their respective MLClassLists allow you to interpret the symbol
names correctly, in light of the corresponding set of definitions.
.. note::
In MUSCIMarker, the MLClassList is currently necessary to define
how CropObjects are displayed: their color. (All noteheads are red,
all barlines are green, etc.) The other function, matching names
to ``clsid``, has been superseeded by the ``clsname`` CropObject
attribute.
**Merging CropObjects**
To merge a list of CropObjects into a new one, you need to:
* Compute the new object's bounding box: ``croobjects_merge_bbox()``
* Compute the new object's mask: ``cropobjects_merge_mask()``
* Determine the clsid and objid of the new object.
Since objid and clsid of merges may depend on external settings
and generally cannot be reliably determined from the merged
objects themselves (e.g. the merge of a notehead and a stem
should be a new note symbol), you need to supply them externally.
However, the bounding box and mask can be determined. The bounding
box is computed simply as the smallest bounding box that
encompasses all the CropObjects, and the mask is an OR operation
over the individual masks (or None, if the CropObjects don't
have masks). Note that the merge cannot deal with a situation
where only some of the objects have a mask.
**Implementation notes on the mask**
The mask is a numpy array that will be saved using run-length encoding.
The numpy array is first flattened, then runs of successive 0's and 1's
are encoded as e.g. ``0:10`` for a run of 10 zeros.
How much space does this take?
Objects tend to be relatively convex, so after flattening, we can expect
more or less two runs per row (flattening is done in ``C`` order). Because
each run takes (approximately) 5 characters, each mask takes roughly ``5 * n_rows``
bytes to encode. This makes it efficient for objects wider than 5 pixels, with
a compression ratio approximately ``n_cols / 5``.
(Also, the numpy array needs to be made C-contiguous for that, which
explains the ``order='C'`` hack in ``set_mask()``.)
"""
def __init__(self,
objid, # type: int
clsname, # type: str
top, # type: int
left, # type: int
width, # type: int
height, # type: int
outlinks=None, # type: Optional[List[int]]
inlinks=None, # type: Optional[List[int]]
mask=None, # type: numpy.ndarray
uid=None, # type: str
data=None
):
# logging.debug('Initializing CropObject with objid {0}, uid {5}, x={1},'
# ' y={2}, h={3}, w={4}'
# ''.format(objid, top, left, height, width, uid))
self.objid = objid
self.clsname = clsname
self.x = top
self.y = left
self.width = width
self.height = height
self.to_integer_bounds()
# The mask presupposes integer bounds.
# Applied relative to CropObject bounds, not the whole image.
self.mask = None
self.set_mask(mask)
if inlinks is None:
inlinks = []
self.inlinks = inlinks
if outlinks is None:
outlinks = []
self.outlinks = outlinks
# Deal with the UID
if uid is None:
uid = self.default_uid
self.set_uid(uid)
self.is_selected = False
# logging.debug('...done!')
if data is None:
data = dict()
self.data = data
##########################################################################
# Dealing with unique identification of a CropObject, also across
# anticipated dataset versions.
UID_DELIMITER = '___'
#: Delimits the CropObject UID fields (global, document namespaces, objid)
UID_DEFAULT_DATASET_NAMESPACE = 'MUSCIMA_DEFAULT_DATASET_PLACEHOLDER'
#: Default dataset name for CropObjects.
UID_DEFAULT_DOCUMENT_NAMESPACE = 'default-document'
#: Default document name for CropObjects.
@property
def default_uid(self):
# type: () -> str
"""Constructs the default ``uid`` that the CropObject would
have, unless one was supplied at initialization.
>>> c.default_uid # doctest: +SKIP
'MUSCIMA_DEFAULT_DATASET_PLACEHOLDER___default-document___0'
"""
return self.UID_DELIMITER.join([self.UID_DEFAULT_DATASET_NAMESPACE,
self.UID_DEFAULT_DOCUMENT_NAMESPACE,
str(self.objid)])
[docs] def parse_uid(self):
# type: () -> (str, str, int)
"""Parse the unique identifier of the CropObject. This
breaks down the UID into the global namespace, document
namespace (ie. CropObjectList name -- usually per image),
and the numeric ID of the CropObject within one CropObjectList.
This numeric ID should always match the ``objid``, which
acts as the "technical" identifier, since it is known to be
an integer and therefore usable for e.g. indexing within
the MUSCIMarker annotation app.
See :meth:`_parse_uid` for format & test. Compared
to :meth:`_parse_uid`, this method checks the parsed ``object_id``
in the ``uid`` against this CropObject's ``objid``,
to verify that the UID is really valid for this object.
The delimiter is expected to be ``___``
(kept as ``CropObject.UID_DELIMITER``)
"""
global_name, document_name, object_id = self._parse_uid(self.uid)
# Dealing with missing uid
if object_id is None:
object_id = self.objid
if object_id != self.objid:
raise ValueError('Got CropObject with different numeric ID'
' in UID and technical objid. UID record:'
' {0}, objid: {1}'.format(object_id, self.objid))
return global_name, document_name, object_id
@staticmethod
def _parse_uid(uid):
# type: (Optional[str]) -> (str, str, int)
"""Parse the unique identifier of the CropObject. This
breaks down the UID into the global namespace, document
namespace (ie. CropObjectList name -- usually per image),
and the numeric ID of the CropObject within one CropObjectList.
The delimiter is expected to be ``___``
(kept as ``CropObject.UID_DELIMITER``)
>>> CropObject._parse_uid('MUSCIMA++_1.0___CVC-MUSCIMA_W-05_N-19_D-ideal___424')
('MUSCIMA++_1.0', 'CVC-MUSCIMA_W-05_N-19_D-ideal', 424)
:returns: ``global_namespace, document_namespace, objid`` triplet.
The namespaces are strings, ``objid`` is an integer. If ``uid``
is ``None``, returns ``None`` as ``objid`` and expects it
to be filled in from the caller CropObject instance.
"""
if uid is None:
global_name = CropObject.UID_DEFAULT_DATASET_NAMESPACE
document_name = CropObject.UID_DEFAULT_DOCUMENT_NAMESPACE
object_id = None
else:
global_name, document_name, numid_str = uid.split(CropObject.UID_DELIMITER)
object_id = int(numid_str)
return global_name, document_name, object_id
[docs] @staticmethod
def build_uid(global_name, document_name, numid):
# type: (Any, Any, Any) -> str
return CropObject.UID_DELIMITER.join([str(global_name),
str(document_name),
str(numid)])
[docs] def set_uid(self, uid):
# type: (str) -> None
"""Assigns the given ``uid`` to the CropObject. This is the way
to do it, do not assign directly to ``cropobject.uid``! You need
to update other things (and perform integrity checks) when changing
the unique ID! See :class:`CropObject` class documentation for
information on how ``uid`` attributes work.
Do **NOT** use this function, unless you know what you are doing!
You could mess up the integrity of your copy of the dataset, and
you'd have to download it again...
"""
self.uid = uid
self._dataset_namespace, self._document_namespace, self._instance = \
self.parse_uid()
[docs] def set_doc(self, docname):
# type: (str) -> None
new_uid = self.UID_DELIMITER.join([self._dataset_namespace,
docname,
str(self._instance)])
self.set_uid(new_uid)
[docs] def set_dataset(self, dataset_name):
# type: (str) -> None
new_uid = self.UID_DELIMITER.join([dataset_name,
self._document_namespace,
str(self._instance)])
self.set_uid(new_uid)
[docs] def set_mask(self, mask):
# type: (numpy.ndarray) -> None
"""Sets the CropObject's mask to the given array. Performs
some compatibilty checks: size, dtype (converts to ``uint8``)."""
if mask is None:
self.mask = None
else:
# Check dimension
t, l, b, r = self.bbox_to_integer_bounds(self.top,
self.left,
self.bottom,
self.right) # .count()
if mask.shape != (b - t, r - l):
raise ValueError('Mask shape {0} does not correspond'
' to integer shape {1} of CropObject.'
''.format(mask.shape, (b - t, r - l)))
if str(mask.dtype) != 'uint8':
logging.debug('CropObject.set_mask(): Supplied non-integer mask'
' with dtype={0}'.format(mask.dtype))
self.mask = mask.astype('uint8')
[docs] def set_objid(self, objid):
# type: (int) -> None
"""Changes the objid and updates the UID with it.
Do NOT use this unless you know what you're doing;
changing the objid should be (1) checked against objid
conflics within the doc, (2) reflected in the outlinks
and inlinks.
"""
self.objid = objid
self._sync_objid_to_uid()
def _sync_objid_to_uid(self):
# type: () -> None
"""Resets the UID number to reflect the objid."""
g_name, doc_name, num = self._parse_uid(self.uid)
new_uid = self.build_uid(g_name, doc_name, self.objid)
self.set_uid(new_uid)
@property
def dataset(self):
# type: () -> str
"""Which dataset is this CropObject coming from?
For bookkeeping."""
# The ``_dataset_namespace`` is set during initialization.
return self._dataset_namespace
@property
def doc(self):
# type: () -> str
"""Which document within the dataset is this CropObject
coming from? The ``_document_namespace``
This is important when working with CropObjects
from multiple CropObjectList files, especially for properly
constructing CropObject graphs, because ``inlinks`` and
``outlinks`` use the numeric ``objids``, which point to
CropObjects within the same document.
``objid`` of each CropObject has to be unique within a document.
"""
# The ``_document_namespace`` is set during initialization.
return self._document_namespace
@property
def top(self):
# type: () -> int
"""Row coordinate of upper left corner."""
return self.x
@property
def bottom(self):
# type: () -> int
"""Row coordinate 1 beyond bottom right corner, so that indexing
in the form ``img[c.top:c.bottom]`` is possible."""
return self.x + self.height
@property
def left(self):
# type: () -> int
"""Column coordinate of upper left corner."""
return self.y
@property
def right(self):
# type: () -> int
"""Column coordinate 1 beyond bottom right corner, so that indexing
in the form ``img[:, c.left:c.right]`` is possible."""
return self.y + self.width
@property
def bounding_box(self):
# type: () -> (int, int, int, int)
"""The ``top, left, bottom, right`` tuple of the CropObject's
coordinates."""
return self.top, self.left, self.bottom, self.right
@property
def middle(self):
# type: () -> (int, int)
"""Returns the integer representation of where the middle
of the CropObject lies, as a ``(m_vert, m_horz)`` tuple.
The integers just get rounded down.
"""
vmid = self.top + (self.bottom - self.top) // 2
hmid = self.left + (self.right - self.left) // 2
return int(vmid), int(hmid)
@property
def is_empty(self):
# type: () -> bool
"""A CropObject is empty if it is composed of zero pixels.
This is measured through the mask. CropObjects without
a mask are assumed to be non-empty."""
if self.mask is None:
return False
return self.mask.sum() == 0
@property
def outlink_uids(self):
# type: () -> List[str]
return [self.build_uid(self.dataset, self.doc, o) for o in self.outlinks]
@property
def inlink_uids(self):
# type: () -> List[str]
return [self.build_uid(self.dataset, self.doc, i) for i in self.inlinks]
[docs] @staticmethod
def bbox_to_integer_bounds(ftop, fleft, fbottom, fright):
# type: (float,float,float,float) -> (int,int,int,int)
"""Rounds off the CropObject bounds to the nearest integer
so that no area is lost (e.g. bottom and right bounds are
rounded up, top and left bounds are rounded down).
Returns the rounded-off integers (top, left, bottom, right)
as integers.
>>> CropObject.bbox_to_integer_bounds(44.2, 18.9, 55.1, 92.99)
(44, 18, 56, 93)
>>> CropObject.bbox_to_integer_bounds(44, 18, 56, 92.99)
(44, 18, 56, 93)
"""
logging.debug('bbox_to_integer_bounds: inputs {0}'.format((ftop, fleft, fbottom, fright)))
top = ftop - (ftop % 1.0)
left = fleft - (fleft % 1.0)
bottom = fbottom - (fbottom % 1.0)
if fbottom % 1.0 != 0:
bottom += 1.0
right = fright - (fright % 1.0)
if fright % 1.0 != 0:
right += 1.0
if top != ftop:
logging.debug('bbox_to_integer_bounds: rounded top by {0}'.format(top - ftop))
if left != fleft:
logging.debug('bbox_to_integer_bounds: rounded left by {0}'.format(left - fleft))
if bottom != fbottom:
logging.debug('bbox_to_integer_bounds: rounded bottom by {0}'.format(bottom - fbottom))
if right != fright:
logging.debug('bbox_to_integer_bounds: rounded right by {0}'.format(right - fright))
return int(top), int(left), int(bottom), int(right)
[docs] def to_integer_bounds(self):
# type: () -> None
"""Ensures that the CropObject has an integer position and size.
(This is important whenever you want to use a mask, and reasonable
whenever you do not need sub-pixel resolution...)
"""
bbox = self.bounding_box
t, l, b, r = self.bbox_to_integer_bounds(*bbox)
height = b - t
width = r - l
self.x = t
self.y = l
self.height = height
self.width = width
[docs] def project_to(self, img):
# type: (numpy.ndarray) -> numpy.ndarray
"""This function returns the *crop* of the input image
corresponding to the CropObject (incl. masking).
Assumes zeros are background."""
# Make a copy! We don't want to modify the original image by the mask.
# Copy forced by the "* 1" part.
crop = img[self.top:self.bottom, self.left:self.right] * 1
if self.mask is not None:
crop *= self.mask
return crop
[docs] def project_on(self, img):
# type: (numpy.ndarray) -> numpy.ndarray
"""This function returns only those parts of the input image
that correspond to the CropObject and masks out everything else
with zeros. The dimension of the returned array is the same
as of the input image. This function basically reconstructs
the symbol as an indicator function over the pixels of
the annotated image."""
output = numpy.zeros(img.shape, img.dtype)
crop = self.project_to(img)
output[self.top:self.bottom, self.left:self.right] = crop
return output
[docs] def render(self, img, alpha=0.3, rgb=(1.0, 0.0, 0.0)):
# type: (numpy.ndarray, float, Tuple[float,float,float]) -> numpy.ndarray
"""Renders itself upon the given image as a rectangle
of the given color and transparency. Might help visualization.
:param img: A three-channel image (3-D numpy array,
with the last dimension being 3)."""
color = numpy.array(rgb)
logging.debug('Rendering object {0}, clsname {1}, t/b/l/r: {2}'
''.format(self.objid, self.clsname,
(self.top, self.bottom, self.left, self.right)))
# logging.debug('Shape: {0}'.format((self.height, self.width, 3)))
mask = numpy.ones((self.height, self.width, 3)) * color
crop = img[self.top:self.bottom, self.left:self.right]
# logging.debug('Mask done, creating crop')
logging.debug('Shape: {0}. Got crop. Crop shape: {1}, img shape: {2}'
''.format((self.height, self.width, 3), crop.shape, img.shape))
mix = (crop + alpha * mask) / (1 + alpha)
img[self.top:self.bottom, self.left:self.right] = mix
return img
[docs] def overlaps(self, bounding_box_or_cropobject):
# type: (Union[Tuple[int,int,int,int],CropObject]) -> bool
"""Check whether this CropObject overlaps the given bounding box or CropObject.
>>> c = CropObject(0, 'test', 10, 100, height=20, width=10)
>>> c.bounding_box
(10, 100, 30, 110)
>>> c.overlaps((10, 100, 30, 110)) # Exact match
True
>>> c.overlaps((0, 100, 8, 110)) # Row mismatch
False
>>> c.overlaps((10, 0, 30, 89)) # Column mismatch
False
>>> c.overlaps((0, 0, 8, 89)) # Total mismatch
False
>>> c.overlaps((9, 99, 31, 111)) # Encompasses CropObject
True
>>> c.overlaps((11, 101, 29, 109)) # Within CropObject
True
>>> c.overlaps((9, 101, 31, 109)) # Encompass horz., within vert.
True
>>> c.overlaps((11, 99, 29, 111)) # Encompasses vert., within horz.
True
>>> c.overlaps((11, 101, 31, 111)) # Corner within: top left
True
>>> c.overlaps((11, 99, 31, 109)) # Corner within: top right
True
>>> c.overlaps((9, 101, 29, 111)) # Corner within: bottom left
True
>>> c.overlaps((9, 99, 29, 109)) # Corner within: bottom right
True
"""
if isinstance(bounding_box_or_cropobject, CropObject):
t, l, b, r = bounding_box_or_cropobject.bounding_box
else:
t, l, b, r = bounding_box_or_cropobject
# Does it overlap vertically? Includes situations where the CropObject is
# inside the bounding box.
# Note that the bottom is +1 (fencepost), so the checks bottom vs. top need to be "less than",
# not leq. If one object's top would be equal to the other's bottom, they would be touching,
# not overlapping.
if max(t, self.top) < min(b, self.bottom):
if max(l, self.left) < min(r, self.right):
return True
return False
[docs] def contains(self, bounding_box_or_cropobject):
"""Check if this CropObject entirely contains the other bounding
box (or, the other cropobject's bounding box)."""
if isinstance(bounding_box_or_cropobject, CropObject):
t, l, b, r = bounding_box_or_cropobject.bounding_box
else:
t, l, b, r = bounding_box_or_cropobject
if self.top <= t <= b <= self.bottom:
if self.left <= l <= r <= self.right:
return True
return False
[docs] def bbox_intersection(self, bounding_box):
# type: (Tuple[int,int, int, int]) -> Optional[Tuple[int,int, int, int]]
"""Returns the sub-bounding box of this CropObject, relative to its size (so: 0,0
is the CropObject's upper left corner), that intersects the given bounding box.
If the intersection is empty, returns None.
>>> c = CropObject(0, 'test', 10, 100, height=20, width=10)
>>> c.bounding_box
(10, 100, 30, 110)
>>> other_bbox = 20, 100, 40, 105
>>> c.bbox_intersection(other_bbox)
(10, 0, 20, 5)
>>> containing_bbox = 4, 55, 44, 115
>>> c.bbox_intersection(containing_bbox)
(0, 0, 20, 10)
>>> contained_bbox = 12, 102, 22, 108
>>> c.bbox_intersection(contained_bbox)
(2, 2, 12, 8)
>>> non_overlapping_bbox = 0, 0, 3, 3
>>> c.bbox_intersection(non_overlapping_bbox) is None
True
"""
t, l, b, r = bounding_box
out_top = max(t, self.top)
out_bottom = min(b, self.bottom)
out_left = max(l, self.left)
out_right = min(r, self.right)
if (out_top < out_bottom) and (out_left < out_right):
return out_top - self.top, \
out_left - self.left, \
out_bottom - self.top, \
out_right - self.left
else:
return None
[docs] def crop_to_mask(self):
# type: () -> None
"""Crops itself to the minimum bounding box that contains all
its pixels, as determined by its mask.
If the mask is all zeros, does not do anything, because
at this point, the is_empty check should be invoked anyway
in any situation where you care whether the object is empty
or not (e.g. delete it after trimming).
>>> mask = numpy.zeros((20, 10))
>>> mask[5:15, 3:8] = 1
>>> c = CropObject(0, 'test', 10, 100, width=10, height=20, mask=mask)
>>> c.bounding_box
(10, 100, 30, 110)
>>> c.crop_to_mask()
>>> c.bounding_box
(15, 103, 25, 108)
>>> c.height, c.width
(10, 5)
Assumes integer bounds, which is ensured during CropObject initialization.
"""
if self.mask is None:
return
if self.is_empty:
return
# We know the object is not empty.
# How many rows/columns to trim from top, bottom, etc.
trim_top = -1
for i in range(self.mask.shape[0]):
if self.mask[i, :].sum() != 0:
trim_top = i
break
trim_left = -1
for j in range(self.mask.shape[1]):
if self.mask[:, j].sum() != 0:
trim_left = j
break
trim_bottom = -1
for k in range(self.mask.shape[0]):
if self.mask[-(k + 1), :].sum() != 0:
trim_bottom = k
break
trim_right = -1
for l in range(self.mask.shape[1]):
if self.mask[:, -(l + 1)].sum() != 0:
trim_right = l
break
logging.debug('Cropobject.crop: Trimming top={0}, left={1},'
'bottom={2}, right={3}'
''.format(trim_top, trim_left, trim_bottom, trim_right))
# new bounding box relative to the current bounding box -- used to trim
# the mask
rel_t = trim_top
rel_l = trim_left
rel_b = self.height - trim_bottom
rel_r = self.width - trim_right
new_mask = self.mask[rel_t:rel_b, rel_l:rel_r] * 1
logging.debug('Cropobject.crop: Old mask shape {0}, new mask shape {1}'
''.format(self.mask.shape, new_mask.shape))
# new bounding box, relative to image -- used to compute the CropObject's
# new position and size
abs_t = self.top + trim_top
abs_l = self.left + trim_left
abs_b = self.bottom - trim_bottom
abs_r = self.right - trim_right
self.x = abs_t
self.y = abs_l
self.height = abs_b - abs_t
self.width = abs_r - abs_l
self.set_mask(new_mask)
def __str__(self):
"""Format the CropObject as its XML representation. See the documentation
of :module:`muscima.io` for details."""
lines = []
lines.append('<CropObject xml:id="{0}">'.format(self.uid))
lines.append('\t<Id>{0}</Id>'.format(self.objid))
# lines.append('\t<UniqueId>{0}</UniqueId>'.format(self.uid))
lines.append('\t<ClassName>{0}</ClassName>'.format(self.clsname))
lines.append('\t<Top>{0}</Top>'.format(self.top))
lines.append('\t<Left>{0}</Left>'.format(self.left))
lines.append('\t<Width>{0}</Width>'.format(self.width))
lines.append('\t<Height>{0}</Height>'.format(self.height))
mask_string = self.encode_mask(self.mask)
lines.append('\t<Mask>{0}</Mask>'.format(mask_string))
if len(self.inlinks) > 0:
inlinks_string = ' '.join(list(map(str, self.inlinks)))
lines.append('\t<Inlinks>{0}</Inlinks>'.format(inlinks_string))
if len(self.outlinks) > 0:
outlinks_string = ' '.join(list(map(str, self.outlinks)))
lines.append('\t<Outlinks>{0}</Outlinks>'.format(outlinks_string))
data_string = self.encode_data(self.data)
if data_string is not None:
lines.append('\t<Data>\n{0}\n\t</Data>'.format(data_string))
lines.append('</CropObject>')
return '\n'.join(lines)
[docs] def encode_mask(self, mask, compress=False, mode='rle'):
# type: (numpy.ndarray, bool, str) -> str
"""Encode a binary array ``mask`` as a string, compliant
with the CropObject format specification in :mod:`muscima.io`.
"""
if mode == 'rle':
return self.encode_mask_rle(mask, compress=compress)
elif mode == 'bitmap':
return self.encode_mask_bitmap(mask, compress=compress)
[docs] def encode_data(self, data):
# type: () -> Optional[str]
if self.data is None:
return None
if len(self.data) == 0:
return None
lines = []
for k, v in list(self.data.items()):
vtype = 'str'
vval = v
if isinstance(v, int):
vtype = 'int'
vval = str(v)
elif isinstance(v, float):
vtype = 'float'
vval = str(v)
elif isinstance(v, list):
vtype = 'list[str]'
if len(v) > 0:
if isinstance(v[0], int):
vtype = 'list[int]'
elif isinstance(v[0], float):
vtype = 'list[float]'
vval = ' '.join([str(vv) for vv in v])
line = '\t\t<DataItem key="{0}" type="{1}">{2}</DataItem>' \
''.format(k, vtype, vval)
lines.append(line)
return '\n'.join(lines)
[docs] def data_display_text(self):
if self.data is None:
return '[No data]'
if len(self.data) == 0:
return '[No data]'
lines = []
for k, v in list(self.data.items()):
lines.append('{0}: {1}'.format(k, v))
return '\n'.join(lines)
[docs] @staticmethod
def encode_mask_bitmap(mask, compress=False):
# type: (numpy.ndarray, bool) -> str
"""Encodes the mask array in a compact form. Returns 'None' if mask
is None. If the mask is not None, uses the following algorithm:
* Flatten the mask (then use width and height of CropObject for
reshaping).
* Record as string, with whitespace separator
* Compress string using gz2 (if compress=True) NOT IMPLEMENTED
* Return resulting string
"""
if mask is None:
return 'None'
# By default works in row-major order.
# So we can just prescribe 'C' without losing data.
mask_flat = mask.flatten(order=CROPOBJECT_MASK_ORDER)
output = ' '.join(list(map(str, mask_flat)))
return output
[docs] @staticmethod
def encode_mask_rle(mask, compress=False):
# type: (numpy.ndarray, bool) -> str
"""Encodes the mask array in Run-Length Encoding. Instead of
having the bitmap ``0 0 1 1 1 0 0 0 1 1``, the RLE encodes
the mask as ``0:2 1:3 0:3 1:2``. This is much more compact.
Currently, the rows of the mask are not treated in any special
way. The mask just gets flattened and then encoded.
Implementation:
"""
if mask is None:
return 'None'
mask_flat = mask.flatten(order=CROPOBJECT_MASK_ORDER)
output_strings = []
current_run_type = 0
current_run_length = 0
for i in mask_flat:
if i == current_run_type:
current_run_length += 1
else:
s = '{0}:{1}'.format(current_run_type, current_run_length)
output_strings.append(s)
current_run_type = i
current_run_length = 1
s = '{0}:{1}'.format(current_run_type, current_run_length)
output_strings.append(s)
output = ' '.join(output_strings)
return output
[docs] def decode_mask(self, mask_string, shape):
# type: (str, Tuple[Any, ...]) -> Optional[numpy.ndarray]
"""Decodes a CropObject mask string into a binary
numpy array of the given shape."""
mode = self._determine_mask_mode(mask_string)
if mode == 'rle':
return self.decode_mask_rle(mask_string, shape=shape)
elif mode == 'bitmap':
return self.decode_mask_bitmap(mask_string, shape=shape)
def _determine_mask_mode(self, mask_string):
# type: (str) -> str
"""If the mask string starts with '0:' or '1:', or generally
if it contains a non-0 or 1 symbol, assume it is RLE."""
mode = 'bitmap'
if len(mask_string) < 3:
mode = 'bitmap'
elif ':' in mask_string[:3]:
mode = 'rle'
return mode
[docs] @staticmethod
def decode_mask_bitmap(mask_string, shape):
# type: (str, Tuple[Any, ...]) -> Optional[numpy.ndarray]
"""Decodes the mask array from the encoded form to the 2D numpy array."""
if mask_string == 'None':
return None
try:
values = list(map(float, mask_string.split()))
except ValueError:
logging.info('CropObject.decode_mask(): Cannot decode mask values:\n{0}'.format(mask_string))
raise
mask = numpy.array(values).reshape(shape)
# s = base64.decodestring(mask_string)
# mask = numpy.frombuffer(s)
# logging.info('CropObject.decode_mask(): shape={0}\nmask={1}'.format(mask.shape, mask))
return mask
[docs] @staticmethod
def decode_mask_rle(mask_string, shape):
# type: (str, Tuple[Any, ...]) -> Optional[numpy.ndarray]
"""Decodes the mask array from the RLE-encoded form
to the 2D numpy array.
"""
if mask_string == 'None':
return None
values = []
for kv in mask_string.split(' '):
k_string, v_string = kv.split(':')
k, v = int(k_string), int(v_string)
vs = [k for _ in range(v)]
values.extend(vs)
mask = numpy.array(values).reshape(shape)
return mask
[docs] def join(self, other):
# type: (CropObject) -> None
"""CropObject "addition": performs an OR on this
and the ``other`` CropObjects' masks and bounding boxes,
and assigns to this CropObject the result. Merges
also the inlinks and outlinks.
Works only if the document spaces for both CropObjects
are the same. (Otherwise changes nothing.)
The ``clsname`` of the ``other`` is ignored.
"""
if self.doc != other.doc:
logging.warning('Trying to join CropObject from doc {0}'
' into this CropObject from doc {1}, skipping.'
''.format(other.doc, self.doc))
return
# Get combined bounding box
nt = min(self.top, other.top)
nl = min(self.left, other.left)
nb = max(self.bottom, other.bottom)
nr = max(self.right, other.right)
nh = nb - nt
nw = nr - nl
# Create mask of corresponding size
new_mask = numpy.zeros((nh, nw), dtype=self.mask.dtype)
# Find coordinates where to paste the masks
spt = self.top - nt # spt = self_paste_top
spl = self.left - nl
opt = other.top - nt
opl = other.left - nl
# Paste the masks into these places
new_mask[spt:spt + self.height, spl:spl + self.width] += self.mask
new_mask[opt:opt + other.height, opl:opl + other.width] += other.mask
# Normalize mask value
new_mask[new_mask != 0] = 1
# Assign the new variables to this CropObject
self.x = nt
self.y = nl
self.height = nh
self.width = nw
self.mask = new_mask
# Add inlinks and outlinks (check for multiple and self-reference)
for o in other.outlinks:
if (o not in self.outlinks) and (o != self.objid):
self.outlinks.append(o)
for i in other.inlinks:
if (i not in self.inlinks) and (i != self.objid):
self.inlinks.append(i)
[docs] def get_outlink_objects(self, cropobjects):
# type: (List[CropObject]) -> List[CropObject]
"""Out of the given ``cropobject`` list, return a list
of those to which this CropObject has outlinks.
Can deal with CropObjects from multiple documents.
"""
output = []
if len(self.outlinks) == 0:
return output
_outlink_set = frozenset(self.outlinks)
for c in cropobjects:
if c.doc != self.doc:
continue
if c.objid in _outlink_set:
output.append(c)
if len(output) == len(self.outlinks):
break
return output
[docs] def get_inlink_objects(self, cropobjects):
# type: (List[CropObject]) -> List[CropObject]
"""Out of the given ``cropobject`` list, return a list
of those from which this CropObject has inlinks.
Can deal with CropObjects from multiple documents.
"""
output = []
if len(self.inlinks) == 0:
return output
_inlink_set = frozenset(self.inlinks)
for c in cropobjects:
if c.doc != self.doc:
continue
if c.objid in _inlink_set:
output.append(c)
if len(output) == len(self.inlinks):
break
return output
[docs] def translate(self, down=0, right=0):
# type: (int, int) -> None
"""Move the cropobject down and right by the given amount of pixels."""
self.x += down
self.y += right
[docs] def scale(self, zoom=1.0):
"""Re-compute the CropObject with the given scaling factor."""
mask = self.mask * 1.0
import skimage.transform
new_mask_shape = max(int(self.height * zoom), 1), max(int(self.width * zoom), 1)
new_mask = skimage.transform.resize(mask,
output_shape=new_mask_shape)
new_mask[new_mask >= 0.5] = 1
new_mask[new_mask < 0.5] = 0
new_mask = new_mask.astype('uint8')
new_height, new_width = new_mask.shape
new_top = int(self.top * zoom)
new_left = int(self.left * zoom)
self.x = new_top
self.y = new_left
self.height = new_height
self.width = new_width
self.mask = new_mask
##############################################################################
# Functions for merging CropObjects and CropObjectLists
[docs]def split_cropobject_on_connected_components(c, next_objid):
# type: (CropObject, int) -> List[CropObject]
"""Split the CropObject into one object per connected component
of the mask. All inlinks/outlinks are retained in all the newly
created CropObjects, and the old object is not changed. (If there
is only one connected component, the object is returned unchanged
in a list of length 1.)
An ``objid`` must be provided at which to start numbering the newly
created CropObjects.
The ``data`` attribute is also retained.
"""
mask = c.mask
# "Safety margin"
canvas = numpy.zeros((mask.shape[0] + 2, mask.shape[1] + 2))
canvas[1:-1, 1:-1] = mask
cc, labels, bboxes = compute_connected_components(canvas)
logging.info('CropObject.split(): {0} ccs, bboxes: {1}'.format(cc, bboxes))
if len(bboxes) == 1:
return [c]
output = []
_next_objid = next_objid
for label, (t, l, b, r) in list(bboxes.items()):
# Background in compute_connected_components() doesn't work?
if label == 0:
continue
h = b - t
w = r - l
m_label = (labels == label).astype('uint8')
m = m_label[t:b, l:r]
top = t + c.top - 1
left = l + c.left - 1
objid = _next_objid
inlinks = copy.deepcopy(c.inlinks)
outlinks = copy.deepcopy(c.outlinks)
data = copy.deepcopy(c.data)
new_c = CropObject(objid, c.clsname, top, left, w, h,
inlinks=inlinks, outlinks=outlinks,
mask=m, data=data)
output.append(new_c)
_next_objid += 1
return output
[docs]def cropobjects_merge(fr, to, clsname, objid):
# type: (CropObject, CropObject, str, int) -> CropObject
"""Merge the given CropObjects with respect to the other.
Returns the new CropObject (without modifying any of the inputs)."""
if fr.doc != to.doc:
raise ValueError('Cannot merge CropObjects from different documents!'
' fr: {0}, to: {1}'.format(fr.doc, to.doc))
mt, ml, mb, mr = cropobjects_merge_bbox([fr, to])
mh = mb - mt
mw = mr - ml
mmask = cropobjects_merge_mask([fr, to])
m_inlinks, m_outlinks = cropobjects_merge_links([fr, to])
m_doc = fr.doc
m_dataset = fr.dataset
m_uid = CropObject.build_uid(m_dataset, m_doc, objid)
output = CropObject(objid, clsname,
top=mt, left=ml, height=mh, width=mw,
mask=mmask,
inlinks=m_inlinks, outlinks=m_outlinks,
uid=m_uid)
return output
[docs]def cropobjects_merge_multiple(cropobjects, clsname, objid):
"""Merge multiple cropobjects. Does not modify any of the inputs."""
if len(set([c.doc for c in cropobjects])) > 1:
raise ValueError('Cannot merge CropObjects from different documents!')
mt, ml, mb, mr = cropobjects_merge_bbox(cropobjects)
mh, mw = mb - mt, mr - ml
m_mask = cropobjects_merge_mask(cropobjects)
m_inlinks, m_outlinks = cropobjects_merge_links(cropobjects)
m_doc = cropobjects[0].doc
m_dataset = cropobjects[0].dataset
m_uid = CropObject.build_uid(m_dataset, m_doc, objid)
output = CropObject(objid, clsname,
top=mt, left=ml, height=mh, width=mw,
mask=m_mask,
inlinks=m_inlinks, outlinks=m_outlinks,
uid=m_uid)
return output
[docs]def cropobjects_merge_bbox(cropobjects):
# type: (List[CropObject]) -> (int, int, int, int)
"""Computes the bounding box of a CropObject that would
result from merging the given list of CropObjects.
"""
# Find extremes. This will define the output cropobject.
t, l, b, r = numpy.inf, numpy.inf, -1, -1
for c in cropobjects:
t = min(t, c.top)
l = min(l, c.left)
b = max(b, c.bottom)
r = max(r, c.right)
it, il, ib, ir = int(t), int(l), int(b), int(r)
if (it != t) or (il != l) or (ib != b) or (ir != r):
logging.warn('Merged bounding box does not consist of integers!'
' {0}'.format((t, l, b, r)))
return it, il, ib, ir
[docs]def cropobjects_merge_mask(cropobjects, intersection=False):
# type: (List[CropObject], bool) -> Optional[numpy.ndarray]
"""Merges the given list of cropobjects into one. Masks are combined
by an OR operation.
>>> c1 = CropObject(0, 'name', 10, 10, 4, 1, mask=numpy.ones((1, 4), dtype='uint8'))
>>> c2 = CropObject(1, 'name', 11, 10, 6, 1, mask=numpy.ones((1, 6), dtype='uint8'))
>>> c3 = CropObject(2, 'name', 9, 14, 2, 4, mask=numpy.ones((4, 2), dtype='uint8'))
>>> c = [c1, c2, c3]
>>> m1 = cropobjects_merge_mask(c)
>>> m1.shape
(4, 6)
>>> print(m1)
[[0 0 0 0 1 1]
[1 1 1 1 1 1]
[1 1 1 1 1 1]
[0 0 0 0 1 1]]
Mask behavior: if at least one of the cropobjects has a mask, then
masking behavior is activated. The masks are combined using OR: any
pixel of the resulting merged cropobject that corresponds to a True
mask pixel in one of the input cropobjects will get a True mask value,
all others (ie. including all intermediate areas) will get a False.
If no input cropobject has a mask, then the resulting cropobject
also will not have a mask.
If some cropobjects have masks and some don't, fails.
:param intersection: Instead of a union, return the mask
intersection: only those pixels which are common to all
the cropobjects.
"""
# No mask
if len([c for c in cropobjects if c.mask is not None]) == 0:
return None
# Some masked, some not
for c in cropobjects:
if c.mask is None:
raise ValueError('Cannot deal with a mix of masked and non-masked cropobjects.')
# Now we know all have masks.
t, l, b, r = cropobjects_merge_bbox(cropobjects)
h = b - t
w = r - l
output_mask = numpy.zeros((h, w), dtype=cropobjects[0].mask.dtype)
# logging.warn('Output mask shape: {0}'.format(output_mask.shape))
for c in cropobjects:
# logging.debug('C. shape: {0}'.format(c.bounding_box))
# logging.debug('TLBR: {0}'.format((t, l, b, r)))
ct, cl, cb, cr = c.top - t, c.left - l, h - (b - c.bottom), w - (r - c.right)
# logging.debug('Mask shape: {0}, curr. shape: {1}'.format(c.mask.shape, (cb - ct, cr - cl)))
output_mask[ct:cb, cl:cr] += c.mask
if intersection:
output_mask[output_mask < len(cropobjects)] = 0
output_mask[output_mask != 0] = 1
else:
output_mask[output_mask > 0] = 1
return output_mask
[docs]def cropobjects_merge_links(cropobjects):
# type: (List[CropObject]) -> (List[CropObject], List[CropObject])
"""Collect all inlinks and outlinks of the given set of CropObjects
to CropObjects outside of this set. The rationale for this is that
these given ``cropobjects`` will be merged into one, so relationships
within the set would become loops and disappear.
(Note that this is not sufficient to update the relationships upon
a merge, because the affected CropObjects *outside* the given set
will need to have their inlinks/outlinks redirected to the new object.)
:returns: A tuple of lists: ``(inlinks, outlinks)``
"""
_internal_objids = frozenset([c.objid for c in cropobjects])
outlinks = []
inlinks = []
for c in cropobjects:
# No duplicates
outlinks.extend([o for o in c.outlinks
if (o not in _internal_objids) and (o not in outlinks)])
inlinks.extend([i for i in c.inlinks
if (i not in _internal_objids) and (i not in inlinks)])
return inlinks, outlinks
[docs]def merge_cropobject_lists(*cropobject_lists):
# type: (List[List[CropObject]]) -> List[CropObject]
"""Combines the CropObject lists from different documents
into one list, so that inlink/outlink references still work.
This is useful only if you want to merge two documents
into one (e.g., if your annotators worked on different "layers"
of data, and you want to merge these annotations).
This just means shifting the ``objid`` (and thus inlinks
and outlinks). It is assumed the lists pertain to the same
image. Uses deepcopy to avoid exposing the original lists
to modification through the merged list.
.. warning::
If you are ever exporting the merged list, make sure to
set the ``uid`` for the outputs correctly, if you want
to create a new document.
.. warning::
Currently cannot handle precedence edges.
"""
max_objids = [max([c.objid for c in c_list]) for c_list in cropobject_lists]
min_objids = [min([c.objid for c in c_list]) for c_list in cropobject_lists]
shift_by = [0] + [sum(max_objids[:i]) - min_objids[i] + 1 for i in range(1, len(max_objids))]
new_lists = []
for clist, s in zip(cropobject_lists, shift_by):
new_list = []
for c in clist:
new_c = copy.deepcopy(c)
# UID handling
collection, doc, _ = new_c.parse_uid()
new_uid = new_c.build_uid(collection, doc, c.objid + s)
new_objid = c.objid + s
new_c.set_uid(new_uid)
new_c.objid = new_objid
# Graph handling
new_c.inlinks = [i + s for i in c.inlinks]
new_c.outlinks = [o + s for o in c.outlinks]
# Should also handle precedence...?
new_list.append(new_c)
new_lists.append(new_list)
output = list(itertools.chain(*new_lists))
return output
[docs]def link_cropobjects(fr, to, check_docname=True):
# type: (CropObject, CropObject, bool) -> None
"""Add a relationship from the ``fr`` CropObject
to the ``to`` CropObject. Modifies the CropObjects
in-place.
If the objects are already linked, does nothing.
:param check_docname: If set, checks for ``docname``
match and raises a ValueError if the CropObjects
come from different documents.
"""
if fr.doc != to.doc:
if check_docname:
raise ValueError('Cannot link two CropObjects that are')
else:
logging.warning('Attempting to link CropObjects from two different'
' docments. From: {0}, to: {1}'
''.format(fr.doc, to.doc))
if (to.objid not in fr.outlinks) and (fr.objid in to.inlinks):
logging.warning('Malformed object graph in document {0}:'
' Relationship {1} --> {2} already exists as inlink,'
' but not as outlink!.'
''.format(fr.doc, fr.objid, to.objid))
fr.outlinks.append(to.objid)
to.inlinks.append(fr.objid)
[docs]def bbox_intersection(bbox_this, bbox_other):
# type: (Tuple[int, int, int, int],Tuple[int, int, int, int]) -> Optional[Tuple[int, int, int, int]]
"""Returns the t, l, b, r coordinates of the sub-bounding box
of bbox_this that is also inside bbox_other.
If the bounding boxes do not overlap, returns None."""
t, l, b, r = bbox_other
tt, tl, tb, tr = bbox_this
out_top = max(t, tt)
out_bottom = min(b, tb)
out_left = max(l, tl)
out_right = min(r, tr)
if (out_top < out_bottom) and (out_left < out_right):
return out_top - tt, \
out_left - tl, \
out_bottom - tt, \
out_right - tl
else:
return None
[docs]def bbox_dice(bbox_this, bbox_other, vertical=False, horizontal=False):
# type: (Tuple[int, int, int, int], Tuple[int, int, int, int], bool, bool) -> float
"""Compute the Dice coefficient (intersection over union)
for the given two bounding boxes.
:param vertical: If set, will only return vertical IoU.
:param horizontal: If set, will only return horizontal IoU.
If both vertical and horizontal are set, will return
normal IoU, as if they were both false.
"""
t_t, t_l, t_b, t_r = bbox_this
o_t, o_l, o_b, o_r = bbox_other
u_t, i_t = min(t_t, o_t), max(t_t, o_t)
u_l, i_l = min(t_l, o_l), max(t_l, o_l)
u_b, i_b = max(t_b, o_b), min(t_b, o_b)
u_r, i_r = max(t_r, o_r), min(t_r, o_r)
u_vertical = max(0, u_b - u_t)
u_horizontal = max(0, u_r - u_l)
i_vertical = max(0, i_b - i_t)
i_horizontal = max(0, i_r - i_l)
if vertical and not horizontal:
if u_vertical == 0:
return 0
else:
return i_vertical / u_vertical
elif horizontal and not vertical:
if u_horizontal == 0:
return 0
else:
return i_horizontal / u_horizontal
else:
if (u_horizontal == 0) or (u_vertical == 0):
return 0
else:
return (i_horizontal * i_vertical) / (u_horizontal * u_vertical)
[docs]def cropobject_distance(c, d):
# type: (CropObject, CropObject) -> numpy.ndarray
"""Computes the distance between two CropObjects.
Their minimum vertical and horizontal distances are each taken
separately, and the euclidean norm is computed from them."""
if c.doc != d.doc:
logging.warning('Cannot compute distances between CropObjects'
' from different documents! ({0} vs. {1})'
''.format(c.doc, d.doc))
c_t, c_l, c_b, c_r = c.bounding_box
d_t, d_l, d_b, d_r = d.bounding_box
delta_vert = 0
delta_horz = 0
if (c_t <= d_t <= c_b) or (d_t <= c_t <= d_b):
delta_vert = 0
elif c_t < d_t:
delta_vert = d_t - c_b
else:
delta_vert = c_t - d_b
if (c_l <= d_l <= c_r) or (d_l <= c_l <= d_r):
delta_horz = 0
elif c_l < d_l:
delta_horz = d_l - c_r
else:
delta_horz = c_l - d_r
return numpy.sqrt(delta_vert ** 2 + delta_horz ** 2)
[docs]def cropobjects_on_canvas(cropobjects, margin=10):
# type: (List[CropObject], int) -> (numpy.ndarray, Tuple[int, int])
"""Draws all the given CropObjects onto a zero background.
The size of the canvas adapts to the CropObjects, with the
given margin.
Also returns the top left corner coordinates w.r.t. CropObjects' bboxes.
"""
# margin is used to avoid the stafflines touching the edges,
# which could perhaps break some assumptions down the line.
it, il, ib, ir = cropobjects_merge_bbox(cropobjects)
_t, _l, _b, _r = max(0, it - margin), max(0, il - margin), ib + margin, ir + margin
canvas = numpy.zeros((_b - _t, _r - _l))
for c in cropobjects:
canvas[c.top - _t:c.bottom - _t, c.left - _l:c.right - _l] = c.mask * 1
canvas[canvas != 0] = 1
return canvas, (_t, _l)
[docs]def cropobject_mask_rpf(cropobject_gt, cropobject_pred):
"""Compute the recall, precision and f-score of the predicted
cropobject's mask against the ground truth cropobject's mask."""
if bbox_intersection(cropobject_gt.bounding_box,
cropobject_pred.bounding_box) is None:
return 0.0, 0.0, 0.0
mask_intersection = cropobjects_merge_mask([cropobject_gt,
cropobject_pred],
intersection=False)
gt_pasted_mask = mask_intersection * 1
t, l, b, r = cropobjects_merge_bbox([cropobject_gt, cropobject_pred])
h, w = b - t, r - l
ct, cl, cb, cr = cropobject_gt.top - t, \
cropobject_gt.left - l, \
h - (b - cropobject_gt.bottom), \
w - (r - cropobject_gt.right)
gt_pasted_mask[ct:cb, cl:cr] += cropobject_gt.mask
gt_pasted_mask[gt_pasted_mask != 0] = 1
pred_pasted_mask = mask_intersection * 1
t, l, b, r = cropobjects_merge_bbox([cropobject_pred, cropobject_pred])
h, w = b - t, r - l
ct, cl, cb, cr = cropobject_pred.top - t, \
cropobject_pred.left - l, \
h - (b - cropobject_pred.bottom), \
w - (r - cropobject_pred.right)
pred_pasted_mask[ct:cb, cl:cr] += cropobject_pred.mask
pred_pasted_mask[pred_pasted_mask != 0] = 1
tp = float(mask_intersection.sum())
fp = pred_pasted_mask.sum() - tp
fn = gt_pasted_mask.sum() - tp
rec, prec = tp / (tp + fn), tp / (tp + fp)
fsc = (2 * rec * prec) / (rec + prec)
return rec, prec, fsc