Today I spent the morning using the Deep Style convolutional neural processing algorithm to create stylized VR landscapes. This processing technique combines the style of one source image with the content of another, as introduced in “A Neural Algorithm of Artistic Style” (2015). A version of this code is available to download on Github. I found a great tutorial on Reddit for using Amazon Web services’ cloud computing to process large files very quickly, which would’ve been perfect for the VR source files I was processing. For this proof of concept, I just used this website to generate low-res DeepStyle images very quickly, making sure to choose art styles that could somewhat survive resampling with bicubic sharpening (pointillism, post-impressionism, cubism).
My source images were VR photos of Laurel Hill Cemetery that I took in the late fall. Laurel Hill is a historic landmark and horticultural garden, so there were plenty of sculptures and beautiful trees to act as the subjects for landscape photography. I filtered them with some very recognizable paintings by Seurat, Van Gogh, and Picasso to experiment with different styles. Overall, this was a very quick and fascinating process with great results, and I’ll likely use this process in the future for speedy matte paintings!
Source: La Grande Jatte by Georges Seurat