gundramleifert / citlabmodule

CITlabModule

This project makes it possible to use the Handwriting Text Recognition (HRT) and the Layout Analysis (LA) from URO/CITlab and Planet i.s. in Transkribus fulfilling the interfaces of TranskribusInterfaces

Building

git clone https://github.com/Transkribus/CITlabModule.git
cd CITlabModule
mvn install

Requirements

general

• java 1.8 or higher
• planet_jar.jar of planet.de (not open available) containing basic libraries

for LA

• OpenCV 3.10

for advanced LA

• OpenCV 3.10
• GCOC (see below)
• tensorflow (see below)

for advanced HTR

• tensorflow (see below)

Building GCOC dependencies for advanced LA

• Ask for the GCOC.tar.gz

tar -xf GCOC.tar.gz
cd GCOC
cmake .
make install

The installation will create HOME/lib and HOME/include (if not existing), we are just interested in the lib file from HOME/lib in case of Linux, it is called libgcoj.so Put this lib somewhere java can find it (LD_LIBRARY_PATH). To add the JNI-Wrapper to your local Maven-Repo, follow:

cd JAVA
sh jar2mvnrepo.sh

Build tensorflow dependencies for advanced LA and HTR:

• the dependency to tensorflow needs a c-lib (an *.so-file)
• the library can be found at https://ci.tensorflow.org/view/Release/job/release-libtensorflow/TYPE=cpu-slave/
• this library should match the version of the java dependency (1.3.0)
• download the library and put the lib somewhere java can find it (LD_LIBRARY_PATH)
• Note that compiling tensorflow on your device can make tensorflow running faster - especially if it can use gpu devices.