docker-ocropy

ocropy in a Docker container.

This is a simple way of getting an ocropy OCR system installed.

Run it

For convenience, this repo comes bundled with a shell script run-ocropy:

> ./run-ocropy --help
./run-ocropy [--help] [--sudo] [--image <image-name>] [--working-dir <working-dir>] <cmd> [args...]

  --sudo                      Run as sudo
  --image <image-name>        Use this image (Default: kbai/ocropy)
  --working-dir <working-dir> Mount this dir (Default: $PWD)
  <cmd>                       Execute this ocropy-<cmd> command
  [args...]                   Depend on the command. See ocropy docs.

Commands:  dewarp econf errs gpageseg gtedit hocr linegen lpred ltrain nlbin rpred rtrain results

Run with docker command

You can run ocropy from the command line:

docker run -v `pwd`:/work kbai/ocropy /ocropy/ocropus-rpred *.png

This will:

Download the docker-ocropy container from Docker Hub
Mount pwd to the /work directory within the docker
Execute /ocropus/ocropy/ocropus-rpred from within the container
On all PNG images in the current dir (pwd) on this computer and hence /work in the container

See the Docker documentation to see how you need to use the -v flag and other flags.

You can also do training and all the other operations you'd normally run with an ocropy installation.

Models

By default, docker-ocropy ships with the models ocropy/en-default and ocropy/fraktur. These are found in /ocropy/models, i.e. use

docker run --rm -it kbai/ocropy ocropus-rpred -m en-default.pyrnn.gz ...
# or
docker run --rm -it kbai/ocropy ocropus-rpred -m /ocropy/models/en-default.pyrnn.gz ...

Train it

Use the ocrotrain.sh script to see how to train the engine:

docker run -v "$PWD:/work" kbai/ocropy ./ocrotrain.sh

This will download the uw3-100 dataset dataset and run ocropus-rtrain on it.

Build it

If you want to adapt the Dockerfile and rebuild the image, you can do so with

docker build [-t '<username>/<reponame>'] .