Read classes from label map

[JR-378]

I grew tired of doing copy and paste from the label map to the script.

There is additional command line parameter to provide the path to the label map. The label map will be read once in the script, parsed into a dictionary and passed to the create_tf_example method.

The names are the keys in the dictionary and the IDs are the values, which makes it work much the same way as the previous method.

All tests passed and tfrecords have been created without issues.

Edit: I recently started training another model and everything seems to be working.

[jarvis-owl]

you could add: import tensorflow.compat.v1 as tf to have tf.app working, if this is ever merged

[JR-378]

Probably never will be. Anyways, my goal was to just share the code, as I am using it in my private repository quite often and without failure.