Fix for unknown command line flags issue

[kristijanbartol]

Fix for missing flags, for example:
"absl.flags._exceptions.UnrecognizedFlagError: Unknown command line flag 'num_source'"

[xiongdemao]

hi！friend，may i have a question about which python version about this project ？

[Yuzz1020]

Hi! I am facing the same problem. Is there any solution to this problem now? The environment I am using is python 2.7.15, tensorflow 1.6.

[tinghuiz]

Hi,
Thanks for the PR! Some comments below.
'num_source' is supposed to be derived from 'seq_length' (i.e. fixed to be seq_length - 1), and 'num_scales' is also fixed to be 4. So neither of them is really supposed to be specified as an input argument. I actually haven't figured out why the current code throws errors on some machines but not on others (e.g. several machines I tested on before the code release). I suspect it's an issue with different tensorflow versions. If you know of the cause, please let me know and I'll see if there's an easy fix.

[JoeFang123]

Maybe you could add two members in FLAGS like this
flags.DEFINE_integer("num_source", None, "number of source images")
flags.DEFINE_integer("num_scales", None, "number of used image scales")
in the train.py