This is an openFrameworks realtime c++ implementation of Logan Engstrom's fast-style-transfer, using the ir image from a kinect as input to the style transfer net. The models are loaded as binary protobuffers using an ugly varhack to preserve the variables in the graph, as explained in ofxMSATensorFlow.
https://github.com/olekristensen/ofFasterStyle/releases/tag/installation
Look at my fork of lengstrom's project and follow the original instructions, then style an image with exactly the same pixel dimensions as the input pixels you want to feed from your ofFasterStyle project and our fork of fast-style-transfer will have saved a of.pb file for you in the model checkpoint dir.
Lilia Amundsen, the master student who initiated this project
DEIC, the super computing guys who let us grind our teeth with this little project on Abacus 2.0
Logan Engstrom, whose tensorflow implementation of a fast style transfer network provided a good starting point for learning.
Memo, who always is ahead of the curve in the openFrameworks community - super awsome addon - really!