You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
ocrd_calamari/ocrd_calamari/recognize.py

270 lines
12 KiB
Python

from __future__ import absolute_import
import os
import itertools
from glob import glob
import numpy as np
from calamari_ocr.ocr import MultiPredictor
from calamari_ocr.ocr.voting import voter_from_proto
from calamari_ocr.proto import VoterParams
from ocrd import Processor
from ocrd_modelfactory import page_from_file
from ocrd_models.ocrd_page import (
LabelType, LabelsType,
MetadataItemType,
TextEquivType,
WordType, GlyphType, CoordsType,
to_xml
)
from ocrd_utils import (
getLogger, concat_padded,
coordinates_for_segment, points_from_polygon, polygon_from_x0y0x1y1,
MIMETYPE_PAGE
)
from ocrd_calamari.config import OCRD_TOOL, TF_CPP_MIN_LOG_LEVEL
TOOL = 'ocrd-calamari-recognize'
log = getLogger('processor.CalamariRecognize')
class CalamariRecognize(Processor):
def __init__(self, *args, **kwargs):
kwargs['ocrd_tool'] = OCRD_TOOL['tools'][TOOL]
kwargs['version'] = OCRD_TOOL['version']
super(CalamariRecognize, self).__init__(*args, **kwargs)
def _init_calamari(self):
os.environ['TF_CPP_MIN_LOG_LEVEL'] = TF_CPP_MIN_LOG_LEVEL
checkpoints = glob(self.parameter['checkpoint'])
self.predictor = MultiPredictor(checkpoints=checkpoints)
voter_params = VoterParams()
voter_params.type = VoterParams.Type.Value(self.parameter['voter'].upper())
self.voter = voter_from_proto(voter_params)
def _make_file_id(self, input_file, n):
file_id = input_file.ID.replace(self.input_file_grp, self.output_file_grp)
if file_id == input_file.ID:
file_id = concat_padded(self.output_file_grp, n)
return file_id
def process(self):
"""
Performs the recognition.
"""
self._init_calamari()
for (n, input_file) in enumerate(self.input_files):
page_id = input_file.pageId or input_file.ID
log.info("INPUT FILE %i / %s", n, page_id)
pcgts = page_from_file(self.workspace.download_file(input_file))
page = pcgts.get_Page()
page_image, page_xywh, page_image_info = self.workspace.image_from_page(page, page_id)
for region in pcgts.get_Page().get_TextRegion():
region_image, region_xywh = self.workspace.image_from_segment(region, page_image, page_xywh)
textlines = region.get_TextLine()
log.info("About to recognize %i lines of region '%s'", len(textlines), region.id)
for (line_no, line) in enumerate(textlines):
log.debug("Recognizing line '%s' in region '%s'", line.id, region.id)
line_image, line_coords = self.workspace.image_from_segment(line, region_image, region_xywh)
line_image_np = np.array(line_image, dtype=np.uint8)
raw_results = list(self.predictor.predict_raw([line_image_np], progress_bar=False))[0]
for i, p in enumerate(raw_results):
p.prediction.id = "fold_{}".format(i)
prediction = self.voter.vote_prediction_result(raw_results)
prediction.id = "voted"
# Build line text on our own
#
# Calamari does whitespace post-processing on prediction.sentence, while it does not do the same
# on prediction.positions. Do it on our own to have consistency.
#
# XXX Check Calamari's built-in post-processing on prediction.sentence
def _sort_chars(p):
"""Filter and sort chars of prediction p"""
chars = p.chars
chars = [c for c in chars if c.char] # XXX Note that omission probabilities are not normalized?!
chars = [c for c in chars if c.probability >= self.parameter['glyph_conf_cutoff']]
chars = sorted(chars, key=lambda k: k.probability, reverse=True)
return chars
def _drop_leading_spaces(positions):
return list(itertools.dropwhile(lambda p: _sort_chars(p)[0].char == " ", positions))
def _drop_trailing_spaces(positions):
return list(reversed(_drop_leading_spaces(reversed(positions))))
def _drop_double_spaces(positions):
def _drop_double_spaces_generator(positions):
last_was_space = False
for p in positions:
if p.chars[0].char == " ":
if not last_was_space:
yield p
last_was_space = True
else:
yield p
last_was_space = False
return list(_drop_double_spaces_generator(positions))
positions = prediction.positions
positions = _drop_leading_spaces(positions)
positions = _drop_trailing_spaces(positions)
positions = _drop_double_spaces(positions)
positions = list(positions)
line_text = ''.join(_sort_chars(p)[0].char for p in positions)
if line_text != prediction.sentence:
log.warning("Our own line text is not the same as Calamari's: '%s' != '%s'",
line_text, prediction.sentence)
# Delete existing results
if line.get_TextEquiv():
log.warning("Line '%s' already contained text results", line.id)
line.set_TextEquiv([])
if line.get_Word():
log.warning("Line '%s' already contained word segmentation", line.id)
line.set_Word([])
# Save line results
line_conf = prediction.avg_char_probability
line.set_TextEquiv([TextEquivType(Unicode=line_text, conf=line_conf)])
# Save word results
#
# Calamari OCR does not provide word positions, so we infer word positions from a. text segmentation
# and b. the glyph positions. This is necessary because the PAGE XML format enforces a strict
# hierarchy of lines > words > glyphs.
def _words(s):
"""Split words based on spaces and include spaces as 'words'"""
spaces = None
word = ''
for c in s:
if c == ' ' and spaces is True:
word += c
elif c != ' ' and spaces is False:
word += c
else:
if word:
yield word
word = c
spaces = (c == ' ')
yield word
if self.parameter['textequiv_level'] in ['word', 'glyph']:
word_no = 0
i = 0
for word_text in _words(line_text):
word_length = len(word_text)
if not all(c == ' ' for c in word_text):
word_positions = positions[i:i+word_length]
word_start = word_positions[0].global_start
word_end = word_positions[-1].global_end
polygon = polygon_from_x0y0x1y1([word_start, 0, word_end, line_image.height])
points = points_from_polygon(coordinates_for_segment(polygon, None, line_coords))
# XXX Crop to line polygon?
word = WordType(id='%s_word%04d' % (line.id, word_no), Coords=CoordsType(points))
word.add_TextEquiv(TextEquivType(Unicode=word_text))
if self.parameter['textequiv_level'] == 'glyph':
for glyph_no, p in enumerate(word_positions):
glyph_start = p.global_start
glyph_end = p.global_end
polygon = polygon_from_x0y0x1y1([glyph_start, 0, glyph_end, line_image.height])
points = points_from_polygon(coordinates_for_segment(polygon, None, line_coords))
glyph = GlyphType(id='%s_glyph%04d' % (word.id, glyph_no), Coords=CoordsType(points))
# Add predictions (= TextEquivs)
char_index_start = 1 # Must start with 1, see https://ocr-d.github.io/page#multiple-textequivs
for char_index, char in enumerate(_sort_chars(p), start=char_index_start):
glyph.add_TextEquiv(TextEquivType(Unicode=char.char, index=char_index, conf=char.probability))
word.add_Glyph(glyph)
line.add_Word(word)
word_no += 1
i += word_length
_page_update_higher_textequiv_levels('line', pcgts)
# Add metadata about this operation and its runtime parameters:
metadata = pcgts.get_Metadata() # ensured by from_file()
metadata.add_MetadataItem(
MetadataItemType(type_="processingStep",
name=self.ocrd_tool['steps'][0],
value=TOOL,
Labels=[LabelsType(
externalModel="ocrd-tool",
externalId="parameters",
Label=[LabelType(type_=name, value=self.parameter[name])
for name in self.parameter.keys()])]))
file_id = self._make_file_id(input_file, n)
pcgts.set_pcGtsId(file_id)
self.workspace.add_file(
ID=file_id,
file_grp=self.output_file_grp,
pageId=input_file.pageId,
mimetype=MIMETYPE_PAGE,
local_filename=os.path.join(self.output_file_grp, file_id + '.xml'),
content=to_xml(pcgts))
# TODO: This is a copy of ocrd_tesserocr's function, and should probably be moved to a ocrd lib
def _page_update_higher_textequiv_levels(level, pcgts):
"""Update the TextEquivs of all PAGE-XML hierarchy levels above `level` for consistency.
Starting with the hierarchy level chosen for processing,
join all first TextEquiv (by the rules governing the respective level)
into TextEquiv of the next higher level, replacing them.
"""
regions = pcgts.get_Page().get_TextRegion()
if level != 'region':
for region in regions:
lines = region.get_TextLine()
if level != 'line':
for line in lines:
words = line.get_Word()
if level != 'word':
for word in words:
glyphs = word.get_Glyph()
word_unicode = u''.join(glyph.get_TextEquiv()[0].Unicode
if glyph.get_TextEquiv()
else u'' for glyph in glyphs)
word.set_TextEquiv(
[TextEquivType(Unicode=word_unicode)]) # remove old
line_unicode = u' '.join(word.get_TextEquiv()[0].Unicode
if word.get_TextEquiv()
else u'' for word in words)
line.set_TextEquiv(
[TextEquivType(Unicode=line_unicode)]) # remove old
region_unicode = u'\n'.join(line.get_TextEquiv()[0].Unicode
if line.get_TextEquiv()
else u'' for line in lines)
region.set_TextEquiv(
[TextEquivType(Unicode=region_unicode)]) # remove old
# vim:tw=120: