Stripped zero-ship samples

... Then I get 35.4% per-image accuracy.

First LGTM full amount on SCC

v1.3

Modify readme

Simply organized version

I don't know it's better or worse, generated with u-net-remote-train.py and more imaages. We will go through a full load on SCC after I have dinner.

Per-image accuracy: 23.6%

The model should work now!

... though with a bad accuracy.

Per-image accuracy: 39.7%

The very first iteration

"Full-res" is not a full-res one (the old one indeed), just trying to make thing up and running.

Files in the release:

• model_fullres_keras.h5

Versioning: We use semantic versioning. That means the major version number will dump if the interface of the model changes.