Philipp Zumstein

Maybe a heroku app for demetsiiify would also be interesting and independent of any server changes?

Yes, they have free option: https://www.heroku.com/pricing and you can AFAIK use docker or docker compose. The databases seems also mentioned in the docu: https://devcenter.heroku.com/categories/data-management . However, I have to admit,...

Maybe, we can add the information about the workshop date then to make it clear for the people visiting here. (We already have some papers form 2011 and before collected.)

See also http://www.businessinsider.de/mit-camera-reads-closed-books-2016-9?r=US&IR=T

Some points from @artunit's talk: - concentrates on OCR of newspapers on microfilms/microfiches - concentrates on ABBYY vs. Tesseract - comparison slide: https://youtu.be/gcjCiS9pJ3A?t=1439 - mentions the Line Segment Detector: http://www.ipol.im/pub/art/2012/gjmr-lsd/...

I think to remember @tmbdev mentions somewhere that one has to train the models for CLSTM again from the GT, i.e. they might not really be convertible.

I would suggest to use `ocropus-econf *.gt.txt` to see the most common confusions, see https://github.com/tmbdev/ocropy/wiki/Compute-errors-and-confusions.

@tmbdev Tom, maybe you can share some insights for this questions. Moreover, I think you are already doing something with tensorflow.

This is IMO quite unrealistic at the moment, but you can try to work with `docker`. There is a [Dockerfile ](https://github.com/tmbdev/clstm/blob/master/Dockerfile)of the project. (@kba is this still up-to-date?)

Shouldn't we add this Dockerfile as well to the repo and maybe rename the current one?