ryanfb/kraken repository overview

⁠kraken-docker

Docker container for the kraken OCR engine⁠.

⁠Usage

Basically, replace kraken in every call with docker run --rm -it ryanfb/kraken kraken, e.g.

kraken --help -> docker run -it ryanfb/kraken kraken --help

Or to drop into an interactive bash shell with the current host directory mapped to the container's /data volume:

docker run -it -v $(pwd):/data ryanfb/kraken /bin/bash

⁠Kraken Models

This container no longer ships with any Kraken models pre-downloaded. There's a persistent Docker volume set up at /models, so you can do e.g.:

docker run -it -v models:/models ryanfb/kraken kraken get default

And downloaded models will persist across runs when you use that volume:

docker run -it -v models:/models ryanfb/kraken ls /models

⁠Shared Memory

If you're trying to train networks with this image, you may run into an error like:

ERROR: Unexpected bus error encountered in worker. This might be caused by insufficient shared memory (shm).

This is because Docker defaults to 64MB of SHM no matter how much memory it's allocated. You can override this by passing e.g. --shm-size=256M as an argument to docker run before the Docker image name.

⁠CUDA

There's also a CUDA variant defined in Dockerfile.cuda that uses the NVIDIA CUDA baseimage and installs a CUDA-enabled pytorch. It's on the cuda tag, and you should be able to run it with nvidia-docker⁠:

nvidia-docker run -it ryanfb/kraken:cuda kraken --help