This image was created using the open source code provided by Google as an initiative of their DeepDream program. DeepDream uses a CNN to find and enhance patterns in images.[1]
We used a digital image of Van Gogh’s Starry Night and trained Google’s code to manipulate a digital image of Taj Mahal. The algorithm runs in iterations to learn from Starry Night and incorporate the artist’s style to redraw Taj Mahal. In short, the code gives an output image of Taj Mahal if it was drawn by Van Gogh!
The scientific interest is in understanding the DeepDream code that works on the basis of Neural Networks and Computer Vision to train the computer. And obtain an image that was produced by learning the artist’s style.
I’d like to thank Yu Zhou, from my lab at CCIPD, CWRU for helping me install Torch, Caffe and getting me started in this beautiful world of deep learning!
References
[1] Mordvintsev, Alexander; Olah, Christopher; Tyka, Mike (2015). “DeepDream – a code example for visualizing Neural Networks”. Google Research.