#!/usr/bin/env python
"""The ``add_staff_relationships.py`` script automates adding
the relationships of some staff-related symbols to staffs.
The symbols in question are:
* ``staff_grouping``
* ``measure_separator``
* ``repeat``
* ``key_signature``
* ``time_signature``
* ``g-clef``, ``c-clef``, ``f-clef``
"""
from __future__ import print_function, unicode_literals, division
import argparse
import collections
import logging
import os
import pprint
import time
from muscima.inference_engine_constants import InferenceEngineConstants as _CONST
from muscima.io import parse_cropobject_list, export_cropobject_list
from muscima.cropobject import link_cropobjects
__version__ = "0.0.1"
__author__ = "Jan Hajic jr."
##############################################################################
ON_STAFFLINE_RATIO_TRHESHOLD = 0.15
STAFFLINE_CLSNAME = 'staff_line'
STAFFSPACE_CLSNAME = 'staff_space'
STAFF_CLSNAME = 'staff'
STAFF_RELATED_CLSNAMES = {
'staff_grouping',
'measure_separator',
'key_signature',
'time_signature',
'g-clef', 'c-clef', 'f-clef', 'other-clef',
}
NOTEHEAD_CLSNAMES = {
'notehead-full',
'notehead-empty',
'grace-notehead-full',
'grace-notehead-empty',
}
REST_CLSNAMES = {
'whole_rest',
'half_rest',
'quarter_rest',
'8th_rest',
'16th_rest',
'32th_rest',
'64th_and_higher_rest',
'multi-measure_rest',
}
# Notes will get added separately.
##############################################################################
[docs]def add_staff_relationships(cropobjects,
notehead_staffspace_threshold=0.2):
_cropobjects_dict = {c.objid: c for c in cropobjects}
ON_STAFFLINE_RATIO_TRHESHOLD = notehead_staffspace_threshold
##########################################################################
logging.info('Find the staff-related symbols')
staffs = [c for c in cropobjects if c.clsname == _CONST.STAFF_CLSNAME]
staff_related_symbols = collections.defaultdict(list)
notehead_symbols = collections.defaultdict(list)
rest_symbols = collections.defaultdict(list)
for c in cropobjects:
if c.clsname in _CONST.STAFF_RELATED_CLSNAMES:
staff_related_symbols[c.clsname].append(c)
if c.clsname in _CONST.NOTEHEAD_CLSNAMES:
notehead_symbols[c.clsname].append(c)
if c.clsname in _CONST.REST_CLSNAMES:
rest_symbols[c.clsname].append(c)
##########################################################################
logging.info('Adding staff relationships')
# - Which direction do the relationships lead in?
# Need to define this.
#
# Staff -> symbol?
# Symbol -> staff?
# It does not really matter, but it's more intuitive to attach symbols
# onto a pre-existing staff. So, symbol -> staff.
for clsname, cs in list(staff_related_symbols.items()):
for c in cs:
# Find the related staff. Relatedness is measured by row overlap.
# That means we have to modify the staff bounding box to lead
# from the leftmost to the rightmost column. This holds
# especially for the staff_grouping symbols.
for s in staffs:
st, sl, sb, sr = s.bounding_box
sl = 0
sr = max(sr, c.right)
if c.overlaps((st, sl, sb, sr)):
link_cropobjects(c, s)
##########################################################################
logging.info('Adding rest --> staff relationships.')
for clsname, cs in list(rest_symbols.items()):
for c in cs:
closest_staff = min([s for s in staffs],
key=lambda x: ((x.bottom + x.top) / 2. - (c.bottom + c.top) / 2.) ** 2)
link_cropobjects(c, closest_staff)
##########################################################################
logging.info('Adding notehead relationships.')
# NOTE:
# This part should NOT rely on staffspace masks in any way!
# They are highly unreliable.
# Sort the staff objects top-down. Assumes stafflines do not cross,
# and that there are no crazy curves at the end that would make the lower
# stafflines stick out over the ones above them...
stafflines = [c for c in cropobjects if c.clsname == _CONST.STAFFLINE_CLSNAME]
stafflines = sorted(stafflines, key=lambda c: c.top)
staffspaces = [c for c in cropobjects if c.clsname == _CONST.STAFFSPACE_CLSNAME]
staffspaces= sorted(staffspaces, key=lambda c: c.top)
staves = [c for c in cropobjects if c.clsname == _CONST.STAFF_CLSNAME]
staves = sorted(staves, key=lambda c: c.top)
# Indexing data structures.
#
# We need to know:
# - per staffline and staffspace: its containing staff
_staff_per_ss_sl = {}
# - per staffline and staffspace: its index (top to bottom) within the staff
_ss_sl_idx_wrt_staff = {}
# Reverse indexes:
# If I know which staff (by objid) and which index of staffline/staffspace,
# I want to retrieve the given staffline/staffspace CropObject:
_staff_and_idx2ss = collections.defaultdict(dict)
_staff_and_idx2sl = collections.defaultdict(dict)
# Build the indexes
for _staff in staves:
# Keep the top-down ordering from above:
_s_stafflines = [_staffline for _staffline in stafflines
if _staff.objid in _staffline.inlinks]
_s_staffspaces = [_staffspace for _staffspace in staffspaces
if _staff.objid in _staffspace.inlinks]
for i, _sl in enumerate(_s_stafflines):
_staff_per_ss_sl[_sl.objid] = _staff
_ss_sl_idx_wrt_staff[_sl.objid] = i
_staff_and_idx2sl[_staff.objid][i] = _sl
logging.debug('Staff {0}: stafflines {1}'.format(_staff.objid,
_staff_and_idx2sl[_staff.objid]))
for i, _ss in enumerate(_s_staffspaces):
_staff_per_ss_sl[_ss.objid] = _staff
_ss_sl_idx_wrt_staff[_ss.objid] = i
_staff_and_idx2ss[_staff.objid][i] = _ss
# pprint.pprint(_ss_sl_idx_wrt_staff)
logging.debug(pprint.pformat(dict(_staff_and_idx2ss)))
# pprint.pprint(dict(_staff_and_idx2sl))
# # Get bounding box of all participating symbols
# notehead_staff_bbox_coords = [c.bounding_box
# for c in list(itertools.chain(*notehead_symbols.values()))
# + stafflines
# + staffspaces
# + staves]
# t, l, b, r = min([b[0] for b in notehead_staff_bbox_coords]), \
# min([b[1] for b in notehead_staff_bbox_coords]), \
# max([b[2] for b in notehead_staff_bbox_coords]), \
# max([b[3] for b in notehead_staff_bbox_coords])
#
# h, w = b - t, r - l
# # Maybe later: ensure a 1-px empty border? (h+2, w+2) and the appropriate +1 when
# # projecting cropobjects onto the canvas...
# staffline_canvas = numpy.zeros((h, w), dtype='uint8')
# for s in stafflines:
# dt, dl, db, dr = s.top - t, s.left - l, s.bottom - t, s.right - l
# staffline_canvas[dt:db, dl:dr] += s.mask
for clsname, cs in list(notehead_symbols.items()):
for c in cs:
ct, cl, cb, cr = c.bounding_box
################
# Add relationship to given staffline or staffspace.
# If notehead has ledger lines, skip it for now.
_has_ledger_line = False
for o in c.outlinks:
if _cropobjects_dict[o].clsname == 'ledger_line':
_has_ledger_line = True
break
if _has_ledger_line:
# Attach to the appropriate staff:
# meaning, staff closest to the innermost ledger line.
lls = [_cropobjects_dict[o] for o in c.outlinks
if _cropobjects_dict[o].clsname == 'ledger_line']
# Furthest from notehead's top is innermost.
# (If notehead is below staff and crosses a ll., one
# of these numbers will be negative. But that doesn't matter.)
ll_max_dist = max(lls, key=lambda ll: ll.top - c.top)
# Find closest staff to max-dist ledger ine
staff_min_dist = min(staves,
key=lambda ss: min((ll_max_dist.bottom - ss.top) ** 2,
(ll_max_dist.top - ss.bottom) ** 2))
link_cropobjects(c, staff_min_dist)
continue
# - Find the related staffline.
# - Because of curved stafflines, this has to be done w.r.t.
# the horizontal position of the notehead.
# - Also, because stafflines are NOT filled in (they do not have
# intersections annotated), it is necessary to use a wider
# window than just the notehead.
# - We will assume that STAFFLINES DO NOT CROSS.
# (That is a reasonable assumption.)
#
# - For now, we only work with more or less straight stafflines.
overlapped_stafflines = []
overlapped_staffline_idxs = []
for i, s in enumerate(stafflines):
# This is the assumption of straight stafflines!
if (ct <= s.top <= cb) or (ct <= s.bottom <= cb):
overlapped_stafflines.append(s)
overlapped_staffline_idxs.append(i)
if c.objid < 10:
logging.debug('Notehead {0} ({1}): overlaps {2} stafflines'.format(c.uid,
c.bounding_box,
len(overlapped_stafflines),
))
if len(overlapped_stafflines) == 1:
s = overlapped_stafflines[0]
dtop = s.top - ct
dbottom = cb - s.bottom
if min(dtop, dbottom) / max(dtop, dbottom) < ON_STAFFLINE_RATIO_TRHESHOLD:
logging.info('Notehead {0}, staffline {1}: very small ratio {2:.2f}'
''.format(c.objid, s.objid, min(dtop, dbottom) / max(dtop, dbottom)))
# Staffspace?
#
# To get staffspace:
# - Get orientation (below? above?)
_is_staffspace_above = False
if dtop > dbottom:
_is_staffspace_above = True
# - Find staffspaces adjacent to the overlapped staffline.
# NOTE: this will fail with single-staffline staves, because
# they do NOT have the surrounding staffspaces defined...
_staffline_idx_wrt_staff = _ss_sl_idx_wrt_staff[s.objid]
if _is_staffspace_above:
_staffspace_idx_wrt_staff = _staffline_idx_wrt_staff
else:
_staffspace_idx_wrt_staff = _staffline_idx_wrt_staff + 1
# Retrieve the given staffsapce
_staff = _staff_per_ss_sl[s.objid]
tgt_staffspace = _staff_and_idx2ss[_staff.objid][_staffspace_idx_wrt_staff]
# Link to staffspace
link_cropobjects(c, tgt_staffspace)
# And link to staff
_c_staff = _staff_per_ss_sl[tgt_staffspace.objid]
link_cropobjects(c, _c_staff)
else:
# Staffline!
link_cropobjects(c, s)
# And staff:
_c_staff = _staff_per_ss_sl[s.objid]
link_cropobjects(c, _c_staff)
elif len(overlapped_stafflines) == 0:
# Staffspace!
# Link to the staffspace with which the notehead has
# greatest vertical overlap.
#
# Interesting corner case:
# Sometimes noteheads "hang out" of the upper/lower
# staffspace, so they are not entirely covered.
overlapped_staffspaces = {}
for _ss_i, s in enumerate(staffspaces):
if s.top <= c.top <= s.bottom:
overlapped_staffspaces[_ss_i] = min(s.bottom, c.bottom) - c.top
elif c.top <= s.top <= c.bottom:
overlapped_staffspaces[_ss_i] = s.bottom - max(c.top, s.top)
if len(overlapped_staffspaces) == 0:
logging.warn('Notehead {0}: no overlapped staffline object, no ledger line!'
''.format(c.uid))
_ss_i_max = max(list(overlapped_staffspaces.keys()),
key=lambda x: overlapped_staffspaces[x])
max_overlap_staffspace = staffspaces[_ss_i_max]
link_cropobjects(c, max_overlap_staffspace)
_c_staff = _staff_per_ss_sl[max_overlap_staffspace.objid]
link_cropobjects(c, _c_staff)
elif len(overlapped_stafflines) == 2:
# Staffspace between those two lines.
s1 = overlapped_stafflines[0]
s2 = overlapped_stafflines[1]
_staff1 = _staff_per_ss_sl[s1.objid]
_staff2 = _staff_per_ss_sl[s2.objid]
if _staff1.objid != _staff2.objid:
raise ValueError('Really weird notehead overlapping two stafflines'
' from two different staves: {0}'.format(c.uid))
_staffspace_idx = _ss_sl_idx_wrt_staff[s2.objid]
s = _staff_and_idx2ss[_staff2.objid][_staffspace_idx]
link_cropobjects(c, s)
# And link to staff:
_c_staff = _staff_per_ss_sl[s.objid]
link_cropobjects(c, _c_staff)
elif len(overlapped_stafflines) > 2:
raise ValueError('Really weird notehead overlapping more than 2 stafflines:'
' {0}'.format(c.uid))
##########################################################################
logging.info('Attaching clefs to stafflines [NOT IMPLEMENTED].')
clefs = [c for c in cropobjects if c.clsname in ['g-clef', 'c-clef', 'f-clef']]
##########################################################################
return cropobjects
###############################################################################
[docs]def build_argument_parser():
parser = argparse.ArgumentParser(description=__doc__, add_help=True,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('-a', '--annot', action='store', required=True,
help='The annotation file for which the staffline and staff'
' CropObject relationships should be added.')
parser.add_argument('-e', '--export', action='store',
help='A filename to which the output CropObjectList'
' should be saved. If not given, will print to'
' stdout.')
parser.add_argument('-t', '--notehead_staffspace_threshold', action='store', type=float,
default=0.2,
help='If the ratio of the smaller to the larger lobe w.r.t.'
' an overlapped staffline is lower than this, we consider'
' the notehead to belong to the adjacent staffspace.')
parser.add_argument('-v', '--verbose', action='store_true',
help='Turn on INFO messages.')
parser.add_argument('--debug', action='store_true',
help='Turn on DEBUG messages.')
return parser
[docs]def main(args):
logging.info('Starting main...')
_start_time = time.clock()
##########################################################################
logging.info('Import the CropObject list')
if not os.path.isfile(args.annot):
raise ValueError('Annotation file {0} not found!'
''.format(args.annot))
cropobjects = parse_cropobject_list(args.annot)
output_cropobjects = add_staff_relationships(
cropobjects,
notehead_staffspace_threshold=args.notehead_staffspace_threshold)
##########################################################################
logging.info('Export the combined list.')
cropobject_string = export_cropobject_list(output_cropobjects)
if args.export is not None:
with open(args.export, 'w') as hdl:
hdl.write(cropobject_string)
else:
print(cropobject_string)
_end_time = time.clock()
logging.info('add_staff_reationships.py done in {0:.3f} s'
''.format(_end_time - _start_time))
if __name__ == '__main__':
parser = build_argument_parser()
args = parser.parse_args()
if args.verbose:
logging.basicConfig(format='%(levelname)s: %(message)s', level=logging.INFO)
if args.debug:
logging.basicConfig(format='%(levelname)s: %(message)s', level=logging.DEBUG)
main(args)