Fix for #8: make forward/backward compatible with tf-1.0.x.

[barryridge]

Fix for #8 based on comments by @vijtad and @zhaozj89.

Add version checks/changes at fail points based on output of tf_upgrade_tool.py.

Add Python 3.x compatibility.

Unsure if adding forward/backward compatibility like this is the best approach for future maintainability. Please advise and/or feel free to reject PR as appropriate.

[warmspringwinds]

@barryridge Thank you so much for your help!

I think tensorflow/models now only works with TF 1.0 or later so, if you can update your PR without backward compatibility I will merge it once I switch to a new version of TF.

[barryridge]

@warmspringwinds You're most welcome! Thank you publishing such a nice package. I'm looking forward to adapting it for another dataset, but I haven't had time yet. I should have time to play with this again next week though, so I'll also look at adjusting the PR then and let you know.

[barryridge]

Apologies for the delay with this! Should be ready for merging now.

[warmspringwinds]

@barryridge Great! Thank you so much! I am glad that you like it.

I will merge once we switch to a new TF version soon.

[barryridge]

@warmspringwinds, FYI, I just noticed while playing around with this stuff yesterday that PR #7 by @ahundt actually incorporates all of these changes anyway and does a lot more besides, so my PR is actually redundant if you merge #7.

[BrianOn99]

@barryridge I think tf_image_segmentation/models/resnet_v1_101_8s.py still have a tf.pack when I locally merged your branch. Should we change it to tf.stack?

[barryridge]

@BrianOn99 Sorry, must have missed that one. Should be fixed now in latest commit.

[BrianOn99]

@barryridge Thanks for the prompt fix.