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

221 lines
5.9 KiB
Plaintext

5 years ago
#!/bin/bash
LOG_LEVEL=${LOG_LEVEL:-DEBUG}
5 years ago
set -e # Abort on error
if [ "$LOG_LEVEL" = "DEBUG" -o "$LOG_LEVEL" = "TRACE" ]; then
set -x
fi
5 years ago
remove_filegrp() {
# Remove the given file group from the workspace
5 years ago
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
5 years ago
}
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-olena-binarize -l $LOG_LEVEL \
-m mets.xml -I OCR-D-IMG -O OCR-D-IMG-BIN \
-p <(echo '{"impl": "sauvola-ms-split"}')
}
do_fontident() {
# Identify fonts in the images
network=`python3 -c "import ocrd_typegroups_classifier, os; print(os.path.join(os.path.dirname(ocrd_typegroups_classifier.__file__), 'models', 'classifier.tgc'))"`
ocrd_typegroups_classifier_parameters="
{
\"network\": \"$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_tesserocr() {
# 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_linesegmentation_sbb() {
# Segment the lines in the images
remove_filegrp OCR-D-SEG-REGION mets.xml
remove_filegrp OCR-D-SEG-LINE mets.xml
ocrd_sbb_textline_detector -l $LOG_LEVEL \
-m mets.xml -I OCR-D-IMG -O OCR-D-SEG-LINE \
-p '{"model": "/var/lib/textline_detection"}'
}
5 years ago
do_ocr() {
# Perform OCR on the segmented lines
ocrd_tesserocr_recognize_parameters='{ "model": "GT4HistOCR_2000000" }' # TODO mods:language + fontident → model
5 years ago
remove_filegrp OCR-D-OCR-TESS mets.xml
ocrd-tesserocr-recognize -l $LOG_LEVEL \
5 years ago
-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
XSD_DIR=`dirname $0`/xsd
if [ ! -d "$XSD_DIR" ]; then
XSD_DIR=/usr/share/xml
fi
xmllint --noout --schema $XSD_DIR/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_fix_image_references_to_bin() {
# Make image references point to the binarized images
# XXX This is a hack, it is probably better to use alternative images in ocrd_calamari
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
# Arrays with filenames to the images
imgs=(`ocrd workspace find -G OCR-D-IMG`)
imgs_bin=(`ocrd workspace find -G OCR-D-IMG-BIN -m image/png`)
# Change all image references to point to the corresponding binarized image
for i in ${!imgs[@]}; do
sed -i "s!imageFilename=.${imgs[$i]}.!imageFilename=\"${imgs_bin[$i]}\"!g" $file
done
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
}
page_upgrade_to_2019() {
filegrp=$1
local file
for file in `ocrd workspace find -G $filegrp`; do
sed -i 's#pagecontent/[0-9-]*#pagecontent/2019-07-15#g' $file
done
}
pip3 list
#do_fontident
#do_validate
do_binarization
do_validate
do_linesegmentation_sbb
page_fix_image_references_to_bin OCR-D-SEG-LINE
page_upgrade_to_2019 OCR-D-SEG-LINE
page_validate_xml OCR-D-SEG-REGION
page_validate_xml OCR-D-SEG-LINE
do_validate
do_ocr_calamari
5 years ago
do_ocr
5 years ago
for ocr_filegrp in OCR-D-OCR-CALAMARI OCR-D-OCR-TESS; do
page_validate_xml $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: