Experimenting with DeepStyle
DeepStyle is a deep learning algorithm similar to Deep Dream based off the research paper "A Neural Algorithm of Artistic Style". The algorithm learns the artistic style of a painter by building filters from features in their artwork, then applies those filters through a constitutional neural network onto a photograph. I used virtualbox to create a VM in which I ran Ubuntu and the program. Unfortunately, I was unable to get the algorithm to run on CUDA, so I had to run it on the CPU instead, greatly increasing run time and memory requirements (bumped my system up to 32GB of RAM for this work). I am going to print some of the better photos on canvas and hang them in my apartment. Generating neural artwork has since grown to become a hobby of mine, and I am constantly working to train the network on new artwork and expand my collection.
Here are some of the paintings, now printed out on canvas: