"""This module implements functions for reading and writing
the data formats used by MUSCIMA++.
Data formats
============
All MUSCIMA++ data is stored as XML, in ``<CropObject>`` elements.
These are grouped into ``<CropObjectList>`` elements, which are
the top-level elements in the ``*.xml`` dataset files.
The list of object classes used in the dataset is also stored as XML,
in ``<CropObjectClass>`` elements (within a ``<CropObjectClassList>``
element).
CropObject
----------
To read a CropObject list file (in this case, a test data file):
>>> from muscima.io import parse_cropobject_list
>>> import os
>>> file = os.path.join(os.path.dirname(__file__), '../test/test_data/01_basic.xml')
>>> cropobjects = parse_cropobject_list(file)
The ``CropObject`` string representation is a XML object::
<CropObject xml:id="MUSCIMA-pp_1.0___CVC-MUSCIMA_W-01_N-10_D-ideal___25">
<Id>25</Id>
<MLClassName>grace-notehead-full</MLClassName>
<Top>119</Top>
<Left>413</Left>
<Width>16</Width>
<Height>6</Height>
<Selected>false</Selected>
<Mask>1:5 0:11 (...) 1:4 0:6 1:5 0:1</Mask>
<Outlinks>12 24 26</Outlinks>
<Inlinks>13</Inlinks>
</CropObject>
The CropObjects are themselves kept as a list::
<CropObjectList>
<CropObjects>
<CropObject> ... </CropObject>
<CropObject> ... </CropObject>
</CropObjects>
</CropObjectList>
Parsing is only implemented for files that consist of a single
``<CropObjectList>``.
Additional information
^^^^^^^^^^^^^^^^^^^^^^
.. caution::
This part may easily be deprecated.
Arbitrary data can be added to the CropObject using the optional
``<Data>`` element. It should encode a dictionary of additional
information about the CropObject that may only apply to a subset
of CropObjects (this facultativeness is what distinguishes the
purpose of the ``<Data>`` element from just subclassing ``CropObject``).
For example, encoding the pitch, duration and precedence information
about a notehead could look like this::
<CropObject>
...
<Data>
<DataItem key="pitch_step" type="str">D</DataItem>
<DataItem key="pitch_modification" type="int">1</DataItem>
<DataItem key="pitch_octave" type="int">4</DataItem>
<DataItem key="midi_pitch_code" type="int">63</DataItem>
<DataItem key="midi_duration" type="int">128</DataItem>
<DataItem key="precedence_inlinks" type="list[int]">23 24 25</DataItem>
<DataItem key="precedence_outlinks" type="list[int]">27</DataItem>
</Data>
</CropObject
The ``CropObject`` will then contain in its ``data`` attribute
the dictionary::
self.data = {'pitch_step': 'D',
'pitch_modification': 1,
'pitch_octave': 4,
'midi_pitch_code': 63,
'midi_pitch_duration': 128,
'precedence_inlinks': [23, 24, 25],
'precedence_outlinks': [27]}
This is also a basic mechanism to allow you to subclass
CropObject with extra attributes without having to re-implement
parsing and export.
.. warning::
Do not misuse this! The ``<Data>`` mechanism is primarily
intended to encode extra information for MUSCIMarker to
display.
Unique identification of a CropObject
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
``xml:id`` is a string that uniquely identifies the CropObject
in the entire dataset. It is derived from a global dataset name and version
identifier (in this case, ``MUSCIMA++_1.0``), a CropObjectList identifier
which is unique within the dataset (derived from the filename:
usually in the format ``CVC-MUSCIMA_W-{:02}_N-{:02}_D-ideal``),
and the number of the CropObject within the given CropObjectList
(which matches the ``<Id>`` value). The delimiter is three underscores
(``___``), in order to comply with XML rules for the ``xml:id`` attribute.
Individual elements of a ``<CropObject>``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
* ``<Id>`` is the integer ID of the CropObject inside a given
``<CropObjectList>`` (which generally corresponds to one XML file
of CropObjects -- one document namespace).
* ``<MLClassName>`` is the name of the object's class (such as
``notehead-full``, ``beam``, ``numeral_3``, etc.).
* ``<Top>`` is the vertical coordinate of the upper left corner of the object's
bounding box.
* ``<Left>`` is the horizontal coordinate of the upper left corner of
the object's bounding box.
* ``<Width>``: the amount of rows that the CropObject spans.
* ``<Height>``: the amount of columns that the CropObject spans.
* ``<Mask>``: a run-length-encoded binary (0/1) 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 not given, the object is understood to occupy the entire
bounding box. For the representation, see Implementation notes
below.
* ``<Inlinks>``: whitespace-separate ``objid`` list, representing CropObjects
**from** which a relationship leads to this CropObject. (Relationships are
directed edges, forming a directed graph of CropObjects.) The objids are
valid in the same scope as the CropObject's ``objid``: don't mix
CropObjects from multiple scopes (e.g., multiple CropObjectLists)!
If you are using CropObjects from multiple CropObjectLists at the same
time, make sure to check against the ``uid``s.
* ``<Outlinks>``: whitespace-separate ``objid`` list, representing CropObjects
**to** which a relationship leads to this CropObject. (Relationships are
directed edges, forming a directed graph of CropObjects.) The objids are
valid in the same scope as the CropObject's ``objid``: don't mix
CropObjects from multiple scopes (e.g., multiple CropObjectLists)!
If you are using CropObjects from multiple CropObjectLists at the same
time, make sure to check against the ``uid``s.
* ``<Data>``: a list of ``<DataItem>`` elements. The elements have
two attributes: ``key``, and ``type``. The ``key`` is what the item
should be called in the ``data`` dict of the loaded CropObject.
The ``type`` attribute encodes the Python type of the item and gets
applied to the text of the ``<DataItem>`` to produce the value.
Currently supported types are ``int``, ``float``, and ``str``,
and ``list[int]``, ``list[float]`` and ``list[str]``. The lists
are whitespace-separated.
The parser function provided for CropObjects does *not* check against
the presence of other elements. You can extend CropObjects for your
own purposes -- but you will have to implement parsing.
Legacy issues with X, Y, and positions
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Formerly, instead of ``<Top>`` and ``<Left>``, there was a different way
of marking CropObject position:
* ``<X>`` was the HORIZONTAL coordinate of the object's upper left corner.
* ``<Y>`` was the VERTICAL coordinate of the object's upper left corner.
Due to legacy issues, the ``<X>`` in the XML file recorded the horizontal
position (column) and ``<Y>`` recorded the vertical position (row). However,
a ``CropObject`` instance uses these attributes in the more natural sense:
``cropobject.x`` is the **top** coordinate, ``cropobject.y`` is the **bottom**
coordinate.
This was unfortunate, and mostly caused by ambiguity of what X and Y mean.
So, the definition of the XML changed: instead of storing nondescript letters,
we will use tags ``<Top>`` and ``<Left>``. Note that we also swapped the order:
where previously the ordering was ``<X>`` (left) first and ``<Y>`` (top)
second, we make ``<Top>`` first and ``<Left>`` second. This corresponds
to how 2-D numpy arrays are indexed: row first, column second.
You may still run into CropObjectList files that use ``<X>`` and ``<Y>``.
The function for reading CropObjectList files, ``parse_cropobject_list()``,
can deal with it and correctly assign the coordinates, but the CropObjects
will be exported with ``<Top>`` and ``<Left>``. (This may break some
there-and-back reencoding tests.)
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 `CROPOBJECT_MASK_ORDER='C'` hack in `set_mask()`.)
CropObjectClass
---------------
This is what a single CropObjectClass element might look like::
<CropObjectClass>
<Id>1</Id>
<Name>notehead-empty</Name>
<GroupName>note-primitive/notehead-empty</GroupName>
<Color>#FF7566</Color>
</CropObjectClass>
See e.g. ``test/test_data/mff-muscima-classes-annot.xml``,
which is incidentally the real CropObjectClass list used
for annotating MUSCIMA++.
Similarly to a ``<CropObjectList>``, the ``<CropObjectClass>``
elements are organized inside a ``<CropObjectClassList>``::
<CropObjectClassList>
<CropObjectClasses>
<CropObjectClass> ... </CropObjectClass>
<CropObjectClass> ... </CropObjectClass>
</CropObjectClasses>
</CropObjectClassesList>
The :class:`CropObjectClass` represents one possible :class:`CropObject`
symbol class, such as a notehead or a time signature. Aside from defining
the "vocabulary" of available object classes for annotation, it also contains
some information about how objects of the given class should
be displayed in the MUSCIMarker annotation software (ordering
related object classes together in menus, implementing a sensible
color scheme, etc.). There is nothing interesting about this class,
we pulled it into the ``muscima`` package because the object
grammar (i.e. which relationships are allowed and which are not)
depends on having CropObjectClass object as its "vocabulary",
and you will probably want to manipulate the data somehow based
on the objects' relationships (like reassembling notes from notation
primitives: notehead plus stem plus flags...), and the grammar
file is a reference for doing that.
"""
from __future__ import print_function, unicode_literals, division
from builtins import str
from builtins import map
import copy
import logging
import os
import collections
from typing import List, Tuple, Optional
from lxml import etree
from muscima.cropobject import CropObject
from muscima.cropobject_class import CropObjectClass
__version__ = "1.0"
__author__ = "Jan Hajic jr."
##############################################################################
[docs]def parse_cropobject_list(filename):
# type: (str) -> List[CropObject]
"""From a xml file with a CropObjectList as the top element, parse
a list of CropObjects. (See ``CropObject`` class documentation
for a description of the XMl format.)
Let's test whether the parsing function works:
>>> test_data_dir = os.path.join(os.path.dirname(os.path.dirname(__file__)),
... 'test', 'test_data',
... 'cropobjects_xy_vs_topleft')
>>> clfile = os.path.join(test_data_dir, '01_basic_topleft.xml')
>>> cropobjects = parse_cropobject_list(clfile)
>>> len(cropobjects)
48
This parsing function can deal with the old-style CropObject XML
that uses ``<Y>`` and ``<X>`` to index the top left corner:
>>> clfile_xy = os.path.join(test_data_dir, '01_basic_xy.xml')
>>> cropobjects_xy = parse_cropobject_list(clfile_xy)
>>> len(cropobjects_xy)
48
>>> len(cropobjects) == len(cropobjects_xy)
True
However, the CropObjects export themselves back only with ``<Top>``
and ``<Left>``.
>>> export_xy = export_cropobject_list(cropobjects_xy)
>>> with open(clfile) as hdl:
... raw_data_topleft = '\\n'.join([l.rstrip() for l in hdl])
>>> raw_data_topleft == export_xy
True
Note that what is Y in the data gets translated to cropobj.x (vertical),
what is X gets translated to cropobj.y (horizontal).
Let's also test the ``data`` attribute:
>>> clfile_data = os.path.join(test_data_dir, '..', '01_basic_binary.xml')
>>> cropobjects = parse_cropobject_list(clfile_data)
>>> cropobjects[0].data['pitch_step']
'G'
>>> cropobjects[0].data['midi_pitch_code']
79
>>> cropobjects[0].data['precedence_outlinks']
[8, 17]
:returns: A list of ``CropObject``s.
"""
tree = etree.parse(filename)
root = tree.getroot()
logging.debug('XML parsed.')
cropobject_list = []
for i, cropobject in enumerate(root.iter('CropObject')):
######################################################
# Parsing one CropObject
logging.debug('Parsing CropObject {0}'.format(i))
# The problem with changing objid: it is often used as
# an integer in MUSCIMarker...
objid = int(float(cropobject.findall('Id')[0].text))
# ...So: we introduce UniqueId (uid). But there needs
# to be some transition, to deal with files that don't
# have it.
try:
try:
# This is ugly, but the xml: namespace is not in the
# cropobject.nsmap dict...
uid = cropobject.attrib['{http://www.w3.org/XML/1998/namespace}id']
except KeyError:
# Fallback
uid = cropobject.findall('UniqueId')[0].text
except:
uid = None
# Generate a default UID at this level?
# No: leave it to the CropObject class default UID mechanism.
# Dealing with clsname transition: there shoud be no more
# CropObjectLists without a clsname. The clsid has been
# removed from CropObject specification, anyway.
# clsname=None
# _has_clsname = False
if len(cropobject.findall('MLClassName')) > 0:
clsname = cropobject.findall('MLClassName')[0].text
elif len(cropobject.findall('ClassName')) > 0:
clsname = cropobject.findall('ClassName')[0].text
else:
raise ValueError('CropObject {0}: no clsname provided.'.format(objid))
#################################
# Top left corner position
# Helper functions for TopLeft/XY transition
def _uses_xy(cropobject):
xs = cropobject.findall('Y')
ys = cropobject.findall('X')
return (len(xs) > 0) and (len(ys) > 0)
def _uses_topleft(cropobject):
xs = cropobject.findall('Top')
ys = cropobject.findall('Left')
return (len(xs) > 0) and (len(ys) > 0)
if _uses_xy(cropobject):
###########################################
# DANGER! DANGER! DANGER! DANGER! DANGER! #
# #
# NOTE THE SWAP OF COORDINATES! #
# #
###########################################
top = float(cropobject.findall('Y')[0].text)
left = float(cropobject.findall('X')[0].text)
elif _uses_topleft(cropobject):
top = float(cropobject.findall('Top')[0].text)
left = float(cropobject.findall('Left')[0].text)
else:
raise KeyError('Cropobject {0} has neither Top/Left, nor Y/Y'
' position information!'.format(objid))
#################################
# Shape
width = float(cropobject.findall('Width')[0].text)
height = float(cropobject.findall('Height')[0].text)
#################################
# Parsing the graph structure (Can deal with missing Inlinks/Outlinks)
inlinks = []
i_s = cropobject.findall('Inlinks')
if len(i_s) > 0:
i_s_text = cropobject.findall('Inlinks')[0].text
if i_s_text is not None: # Zero-length links
inlinks = list(map(int, i_s_text.split(' ')))
outlinks = []
o_s = cropobject.findall('Outlinks')
if len(o_s) > 0:
o_s_text = cropobject.findall('Outlinks')[0].text
if o_s_text is not None:
outlinks = list(map(int, o_s_text.split(' ')))
#################################
# Decode the data.
data = cropobject.findall('Data')
data_dict = None
if len(data) > 0:
data = data[0]
data_dict = {}
for data_item in data.findall('DataItem'):
key = data_item.get('key')
value_type = data_item.get('type')
value = data_item.text
# logging.debug('Creating data entry: key={0}, type={1},'
# ' value={2}'.format(key, value_type, value))
if value_type == 'int':
value = int(value)
elif value_type == 'float':
value = float(value)
elif value_type.startswith('list'):
if value is None:
value = []
else:
vt_factory = str
if value_type.endswith('[int]'):
vt_factory = int
elif value_type.endswith('[float]'):
vt_factory = float
value = list(map(vt_factory, value.split()))
data_dict[key] = value
#################################
# Create the object.
obj = CropObject(objid=objid,
uid=uid,
clsname=clsname,
top=top,
left=left,
width=width,
height=height,
inlinks=inlinks,
outlinks=outlinks,
data=data_dict)
#################################
# Add mask.
# We do this only after the CropObject has been created,
# to make sure that the width & height used to reshape
# the flattened mask reflects what is in the CropObject.
mask = None
m = cropobject.findall('Mask')
if len(m) > 0:
mask = obj.decode_mask(cropobject.findall('Mask')[0].text,
shape=(obj.height, obj.width))
obj.set_mask(mask)
# logging.debug('Created CropObject with ID {0}'.format(obj.objid))
cropobject_list.append(obj)
logging.debug('CropObjectList loaded.')
if not validate_cropobjects_graph_structure(cropobject_list):
raise ValueError('Invalid CropObject graph structure! Check warnings'
' in log for the individual errors.')
return cropobject_list
[docs]def validate_cropobjects_graph_structure(cropobjects):
# type: (List[CropObject]) -> bool
"""Check that the graph defined by the ``inlinks`` and ``outlinks``
in the given list of CropObjects is valid: no relationships
leading from or to objects with non-existent ``objid``s.
Can deal with ``cropobjects`` coming from a combination
of documents, through the CropObject ``doc`` property.
Warns about documents which are found inconsistent.
:param cropobjects: A list of :class:`CropObject` instances.
:returns: ``True`` if graph is valid, ``False`` otherwise.
"""
# Split into lists by doc
cropobjects_by_doc = collections.defaultdict(list)
for c in cropobjects:
cropobjects_by_doc[c.doc].append(c)
is_valid = True
for doc, doc_cropobjects in list(cropobjects_by_doc.items()):
doc_is_valid = validate_document_graph_structure(doc_cropobjects)
if not doc_is_valid:
logging.warning('Document {0} has invalid cropobject graph!'
''.format(doc))
is_valid = False
return is_valid
[docs]def validate_document_graph_structure(cropobjects):
# type: (List[CropObject]) -> bool
"""Check that the graph defined by the ``inlinks`` and ``outlinks``
in the given list of CropObjects is valid: no relationships
leading from or to objects with non-existent ``objid``s.
Checks that all the CropObjects come from one document. (Raises
a ``ValueError`` otherwise.)
:param cropobjects: A list of :class:`CropObject` instances.
:returns: ``True`` if graph is valid, ``False`` otherwise.
"""
docs = [c.doc for c in cropobjects]
if len(set(docs)) != 1:
raise ValueError('Got CropObjects from multiple documents!')
is_valid = True
objids = frozenset([c.objid for c in cropobjects])
for c in cropobjects:
inlinks = c.inlinks
for i in inlinks:
if i not in objids:
logging.warning('Invalid graph structure in CropObjectList:'
' object {0} has inlink from non-existent'
' object {1}'.format(c, i))
is_valid = False
outlinks = c.outlinks
for o in outlinks:
if o not in objids:
logging.warning('Invalid graph structure in CropObjectList:'
' object {0} has outlink to non-existent'
' object {1}'.format(c, o))
is_valid = False
return is_valid
[docs]def export_cropobject_graph(cropobjects, validate=True):
# type: (List[CropObject], bool) -> List[Tuple[int, int]]
"""Collects the inlink/outlink CropObject graph
and returns it as a list of ``(from, to)`` edges.
:param cropobjects: A list of CropObject instances.
All are expected to be within one document.
:param validate: If set, will raise a ValueError
if the graph defined by the CropObjects is
invalid.
:returns: A list of ``(from, to)`` objid pairs
that represent edges in the CropObject graph.
"""
if validate:
validate_cropobjects_graph_structure(cropobjects)
edges = []
for c in cropobjects:
for o in c.outlinks:
edges.append((c.objid, o))
return edges
[docs]def export_cropobject_list(cropobjects, docname=None, dataset_name=None):
# type: (List[CropObject], Optional[str], Optional[str]) -> str
"""Writes the CropObject data as a XML string. Does not write
to a file -- use ``with open(output_file) as out_stream:`` etc.
:param cropobjects: A list of CropObject instances.
:param docname: Set the document name for all the CropObject
unique IDs to this. If not given, no docname is applied.
This means that either the old document identification
stays (in case the CropObjects are loaded from a file
with document IDs set), or the default is used (if the
CropObjects have been newly created). If given,
the CropObjects are first deep-copied, so that the existing
objects' UID is not affected by the export.
:param dataset_name: Analogous to docname.
"""
if docname is not None:
new_cropobjects = []
for c in cropobjects:
new_c = copy.deepcopy(c)
new_c.set_doc(docname)
new_cropobjects.append(new_c)
cropobjects = new_cropobjects
if dataset_name is not None:
new_cropobjects = []
for c in cropobjects:
new_c = copy.deepcopy(c)
new_c.set_dataset(dataset_name)
new_cropobjects.append(new_c)
cropobjects = new_cropobjects
# This is the data string, the rest is formalities
cropobj_string = '\n'.join([str(c) for c in cropobjects])
lines = list()
lines.append('<?xml version="1.0" encoding="utf-8"?>')
lines.append('<CropObjectList'
' xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"'
' xmlns:xsd="http://www.w3.org/2001/XMLSchema">')
lines.append('<CropObjects>')
lines.append(cropobj_string)
lines.append('</CropObjects>')
lines.append('</CropObjectList>')
return '\n'.join(lines)
##############################################################################
# Parsing CropObjectClass lists, mostly for grammars.
[docs]def parse_cropobject_class_list(filename):
# type: (str) -> List[CropObjectClass]
"""From a xml file with a MLClassList as the top element,
extract the list of :class:`CropObjectClass` objects. Use
this
"""
tree = etree.parse(filename)
root = tree.getroot()
cropobject_class_list = []
for mlclass in root.iter('CropObjectClass'):
cls = CropObjectClass(clsid=int(mlclass.findall('Id')[0].text),
name=mlclass.findall('Name')[0].text,
group_name=mlclass.findall('GroupName')[0].text,
color=mlclass.findall('Color')[0].text)
cropobject_class_list.append(cls)
return cropobject_class_list
[docs]def export_cropobject_class_list(cropobject_classes):
# type: (List[CropObjectClass]) -> str
"""Writes the CropObject data as a XML string. Does not write
to a file -- use ``with open(output_file) as out_stream:`` etc.
:param cropobject_classes: A list of CropObjectClass instances.
"""
cropobject_classes_string = '\n'.join([str(c) for c in cropobject_classes])
lines = list()
lines.append('<?xml version="1.0" encoding="utf-8"?>')
lines.append('<CropObjectClassList'
' noNamespaceSchema="mff-muscima-cropobject-classes.xsd"'
' xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"'
' xmlns:xsd="http://www.w3.org/2001/XMLSchema">')
lines.append('<CropObjectClasses>')
lines.append(cropobject_classes_string)
lines.append('</CropObjectClasses>')
lines.append('</CropObjectClassList>')
return '\n'.join(lines)