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-galley/my_ocrd_workflow

187 lines
5.0 KiB
Bash

#!/bin/bash
LOG_LEVEL=${LOG_LEVEL:-DEBUG}
set -e # Abort on error
if [ "$LOG_LEVEL" = "DEBUG" -o "$LOG_LEVEL" = "TRACE" ]; then
set -x
fi
remove_filegrp() {
# Remove the given file group from the workspace
filegrp_use=$1
mets=$2
xmlstarlet ed --inplace \
-N mets=http://www.loc.gov/METS/ \
-d "//mets:fileGrp[@USE='$filegrp_use']" $mets
# XXX See also https://github.com/OCR-D/core/issues/245
# XXX This should also delete the files (after checking if they are indeed inside the workspace) and the directory
}
do_validate() {
ocrd workspace validate --skip pixel_density --page-strictness lax mets.xml
# XXX ocrd-tesserocr INCONSISTENCY in TextRegion → use "--page-strictness lax" for now
# XXX
if test -d TEMP; then
echo "TEMP exists!"
rm -rf TEMP
fi
}
do_binarization() {
# Binarize the images
remove_filegrp OCR-D-IMG-BIN mets.xml
ocrd-kraken-binarize -l $LOG_LEVEL \
-m mets.xml -I OCR-D-IMG -O OCR-D-IMG-BIN
}
do_fontident() {
# Identify fonts in the images
network=`python3 -c "import ocrd_typegroups_classifier, os; print(os.path.join(os.path.dirname(ocrd_typegroups_classifier.__file__), 'models', 'classifier.tgc'))"`
ocrd_typegroups_classifier_parameters="
{
\"network\": \"$network\",
\"stride\": 143
}"
remove_filegrp OCR-D-OCR-FONTIDENT mets.xml
ocrd-typegroups-classifier -l $LOG_LEVEL \
-m mets.xml -I OCR-D-IMG -O OCR-D-OCR-FONTIDENT \
-p <(echo $ocrd_typegroups_classifier_parameters)
# XXX Check if ocrd-typegroups-classifier uses the whole image
# XXX does DEFAULT have any meaning? /buerger_gedichte_1778.ocrd does not have
# any DEFAULT, yet -I DEFAULT seems to work for ocrd-typegroups-classifier
}
do_linesegmentation() {
# Segment the lines in the binarized images
remove_filegrp OCR-D-SEG-REGION mets.xml
remove_filegrp OCR-D-SEG-LINE mets.xml
#ocrd-ocropy-segment -l $LOG_LEVEL \
# -m mets.xml -I OCR-D-IMG-BIN -O OCR-D-SEG-LINE
# XXX ocrd-ocropy-segment throws an exception for buerger_gedichte_1778.ocrd
ocrd-tesserocr-segment-region -l $LOG_LEVEL \
-m mets.xml -I OCR-D-IMG-BIN -O OCR-D-SEG-REGION
ocrd-tesserocr-segment-line -l $LOG_LEVEL \
-m mets.xml -I OCR-D-SEG-REGION -O OCR-D-SEG-LINE
# XXX compare ocrd-tesserocr-segment* vs tesseract native
}
do_ocr() {
# Perform OCR on the segmented lines
ocrd_tesserocr_recognize_parameters='{ "model": "frk" }' # TODO mods:language + fontident → model
remove_filegrp OCR-D-OCR-TESS mets.xml
ocrd-tesserocr-recognize -l $LOG_LEVEL \
-m mets.xml -I OCR-D-SEG-LINE -O OCR-D-OCR-TESS \
-p <(echo $ocrd_tesserocr_recognize_parameters)
}
do_ocr_calamari() {
ocrd_calamari_recognize_parameters='{ "checkpoint": "/var/lib/calamari-models/GT4HistOCR/*.ckpt.json" }'
remove_filegrp OCR-D-OCR-CALAMARI mets.xml
ocrd-calamari-recognize -l $LOG_LEVEL \
-m mets.xml -I OCR-D-SEG-LINE -O OCR-D-OCR-CALAMARI \
-p <(echo $ocrd_calamari_recognize_parameters)
}
page_validate_xml() {
# Validate all PAGE XML against the XML schema
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
xmllint --noout --schema `dirname $0`/xsd/pagecontent.2019-07-15.xsd $file
done
}
page_fix_image_references() {
# Make image references relative to the PAGE XML file
#
# The rest of OCR-D probably isn't going to like it, but it is a. correct and b. makes PAGE Viewer open the image file
# automatically.
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
sed -i 's#imageFilename="OCR-D-IMG#imageFilename="../OCR-D-IMG#g' $file
done
}
page_workaround_remove_conf() {
# XXX Work around https://github.com/OCR-D/core/issues/269
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
xmlstarlet ed --inplace \
-N 'page=http://schema.primaresearch.org/PAGE/gts/pagecontent/2019-07-15' \
-d '//page:TextEquiv/@conf' $file
done
}
page_downgrade_to_2018() {
# Not used anymore, but kept if needed in the future
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
sed -i 's#pagecontent/[0-9-]*#pagecontent/2018-07-15#g' $file
done
}
# FIXME We have a problem with ocrd_typegroups_classifier's Pillow dependency atm.
#do_fontident
#do_validate
do_binarization
do_validate
do_linesegmentation
page_validate_xml OCR-D-SEG-REGION
page_validate_xml OCR-D-SEG-LINE
do_validate
do_ocr_calamari
do_ocr
for ocr_filegrp in OCR-D-OCR-CALAMARI OCR-D-OCR-TESS; do
page_validate_xml $ocr_filegrp
page_workaround_remove_conf $ocr_filegrp
do_validate
page_validate_xml $ocr_filegrp
do_validate
# XXX This seems to be causing new problems with validation
# https://github.com/OCR-D/core/issues/176
#page_fix_image_references $ocr_filegrp
#do_validate
if ocrd workspace list-group | grep -q OCR-D-GT-PAGE; then
remove_filegrp $ocr_filegrp-EVAL mets.xml
ocrd-dinglehopper -m mets.xml -I OCR-D-GT-PAGE,$ocr_filegrp -O $ocr_filegrp-EVAL
fi
done
# vim:tw=120: