This PR adds a new extractors.models submodule that will house Extractors based on pre-trained DNN models. I've added a new TensorFlowInceptionV3Extractor that does image labeling based on the pre-trained Inception V3 model discussed in the TF docs. See test for example usage, which follows the standard Extractor pattern. The 200 MB pretrained model will be downloaded the first time the extractor is run, if it' s not found, and no additional code is required.
added new TensorFlowInceptionV3Extractor based on TF example
update travis to install TensorFlow so new test doesn't break
update setuptools before pip install to fix TF-related issue
switch travis-ci to trusty distribution to try to solve TF build errors
ffmpeg was removed from trusty, so install it from mc3man ppa
try installing from a different ppa
replace urllib with requests in _download_pretrained_models() for Py …
kludgy workaround for moviepy buffer size bug
Coverage increased (+0.6%) to 79.567% when pulling 0ee9163 on inception-v3 into 1f8d384 on master.