Ocropus data to TF-CRNN implementation

This small script is meant to be used if you have some manuscripts data and you want to try out Sofia Ares Oliveira's implementation of TF-CRNN ( https://github.com/solivr/tf-crnn for more details)

Use

This script is a single file script at the moment and we hope to keep it this way. Download or clone the directory beforehand. You can also simply copy the content of cli.py and save it where you need it.

To use it, you can do :

python3 cli.py my_ocropus_data_directory --output tf-crnn

This will create a file "groundtruth.csv" in the directory ./tf-crnn/. The image file links are absolute path, which allows to move your groundtruth file to be moved along.

The command accepts also multiple directory and unix-selectors for the input directories.

python3 cli.py ocropus/train  ocropus/test --output tf-crnn
# Same as
python3 cli.py ocropus/* --output tf-crnn

Both these commands will create files in tf-crnn/train and tf-crnn/test/ named groundtruth.csv/.