Joel Akeret

We formulated the problem as a 3-class classification and used a numerical simulation to generate labelled training ground truth data

`tf_unet` stores the intermediate model checkpoints under the `output_path`. You should be able to reload the tf graph from there

Sounds about right

Hmm thats something I haven't seen before

Sweet! Thanks very much for digging into this!

Great to hear that the package is useful to you. Any kind of contribution is highly welcomed. In principle I don't see a problem. Ideally, the padding becomes a parameter...

For the maxpooling the padding shouldn't have an impact. For the conv/deconv this could indeed introduce a bias but i'm not sure how big the impact would be

Sweet that looks nice. I will play arround a little bit to get a feelling but at first sight this looks good. Thanks a lot!

@ethanbb I looked at your PR and quickly ran into some issues. I was using the `launcher.py` example. Any thoughts? > 2017-06-07 21:41:40,671 Layers 3, features 16, filter size 3x3,...

First of all thanks for your contribution. I just found some time to have a look. I noticed that there are [syntax errors](https://github.com/jakeret/tf_unet/pull/51/files#diff-58534af1919b4a14e4883a69fc86e45fR239) in the commit. When I fixed them...