Improve documentation for dewarp

[VolkerHartmann]

Requirements:
Installation of pix2pixHD
pip3 install torch torchvision
pip install dominate
parameter model_name seems to be ignored
set parameter for pix2pixHD

[VolkerHartmann]

Still got an error:
File "/home/user/ocrd/python/pix2pixHD/models/pix2pixHD_model.py", line 58, in initialize
self.load_network(self.netG, 'G', opt.which_epoch, pretrained_path)
File "/home/user/ocrd/python/pix2pixHD/models/base_model.py", line 58, in load_network
raise('Generator must exist!')

Please provide an example how to configure dewarper correct.
Also a link inside the README.md to the second README.md would be helpful.

[VolkerHartmann]

typo in ocrd-anybaseocr-block-segmenter -> ocrd-anybaseocr-block-segmentation

[bertsky]

The packaging of this module (and its documentation) is a nightmare:

Installation of pix2pixHD

Since there's no release on PyPI, this literally means cloning https://github.com/NVIDIA/pix2pixHD
somewhere, and creating a checkpoints subdirectory for it, and then passing the absolute pathname of the clone to the pix2pixHD parameter

pip3 install torch torchvision
pip install dominate

along with a full CUDA setup

parameter model_name seems to be ignored

No, it's not. It needs to point to the absolute pathname of the directory holding the dewarping model manually downloaded from here

All this should have been encapsulated in requirements.txt or at least in some Makefile recipe for installation. Since there's no distribution for it, the dependency on pix2pixHD should be made a submodule. The requirement to give absolute path names (twice!) in the parameter JSON makes it impossible to ship host-independent parameter/workflow configurations. IMO the pix2pixHD path should not be a parameter at all, but resolved at install-time.

[kba]

fixed by #89