Neural-Style

I have been turned on to a project called Neural Style. I'm terribly familiar with or interested in neural nets. There has been a lot of media exposure lately on the subject and I haven't read any of it. What I have done is stumble upon an amusing imgur post demonstrating what can be done by using a neural net to re-imagine an image (the content image) in the style of another source image (style image). The software wrapper to accomplish this is Neural Style. The software is easy enough to use; if not to install and configure. The software it depends on isn't exactly a prime application of Unix design principles. The various tools and libraries appear to be in a constant state of feature and API flux. But no matter. For quite a while I used only CPU power and system core memory to process these, since the video card I had at first has only 1GB of memory. OpenCL (and CUDA) support are available, but as of now can only address memory in the GPU. AMD has big talk and big plans with regards to their APU work that they will soon have a seamless integration between main and gpu memory for scientific computing purposes. We will be pleased if this bears fruit. Later renders are done with GPU acceleration since I bought an RX480 8GB video card.

Here is a selection of my favorites; they will be featured in the order of content image, result, style.

This is one of the first renders I did. Featured on the left is me sleeping on my cat, Dixie. For the style we have Edward Gorey.

Some stock images of cows and tile floors.

I like how this one turned out. The result basically what I was expected to happen.

This one always makes me giggle.

Steve Buscemi. 'Nuff said.

The old Indiana Theatre, now the Buskirk-Chumley combined with a screen grab from Jet Set Radio.

And the Kremlin recieves the same Jet Set treatment.

World of Warcraft fans will understand a Molten Corgi.