* initial public refactor

* add uniform

* simplify code

* split uniform-factor

* fix prior

* working version

* more documentation

* clean up done for InfoGAN

* updated other examples to match changes in GAN.py

* final edits from my side

* docs changes in base class

* more docs and interface name

* some changes in name scope and docs

* use property for _dim

* visualization & gradient scale according to the paper

* update the demo img

* Add example for several embedding methods

This also includes an animate script using avconv and app-util to
reduce repeating code (DRY).

* some docs

* use average for cosine-loss. add some math.

* move data to embedding_data.py; fix inference

* simplify data code

* Change rnn-cell to fix #103

* Add example for extracting saliency using ResNet-50 with tfSlim

* use existing mean. document the symbolic functions

* move function out of utils.py

* further simplify code. move saliency back to symbolic_functions.

* rename the script. add stacked demo image.

* remove cat.jpg