Ever wondered how Picasso could have painted The Mona Lisa?

I used the ideas of Neural Style Transfer presented by Andrew Ng in his deep learning course to optimize hidden layer activations of a pretrained Convolutional Neural Network (ConvNet) to generate one!

The algorithm was fed the image of Mona Lisa painting as well as a second “style image” representing Picasso’s painting style. I then used Tensorflow to optimize the ConvNet in 200 iterations. Here is the input and output.

The following video shows the generated image being progressively built (intermediate results are rendered every 10 iterations.)