[WIP] Image captioning

[ebenolson]

Work in progress on an image caption generation example, based on https://github.com/karpathy/neuraltalk

I'm new to RNN, so please let me know if I'm doing anything especially silly. In particular I'm sure there's a better way to generate samples than I'm doing now.

TODO:

• [ ] More background and explanatory comments
• [ ] Add a section or new notebook for calculating BLEU score
• [ ] Tune parameters better
• [ ] More efficient sampling / add beam search
• [ ] Region model (http://cs.stanford.edu/people/karpathy/cvpr2015.pdf) - probably a separate PR.

[f0k]

(Note, just so it's not overlooked: That PR is from your master branch, you may either want to file a new PR from a separate branch or rebase it onto upstream/master before merging. Right now you've accidentally included an art style transfer commit in there.)