Art, By A Neural Network

Udacity’s Deep Learning Nanodegree foundations course has finally begun.

One of the first things we got to do was make art with a neural network.

Using Logan Engstrom’s Fast Style Transfer, I input different pictures into a trained neural network. The network then output that picture, but made in the style of a famous artist.

For example, The fast style transfer network was trained on this painting, “A Muse” by Pablo Picasso:

I then provided the neural net with this picture of my girlfriend and me:

And the result is:

Admittedly, I do look a bit strange there.

I tried another. With the neural network trained on “The Scream”…

…I got this painting of my friends and me:

“That’s cool, but don’t phones already do this? What’s so life-changing about this?”

That was my first question when I heard about neural nets creating this type of art.

But after playing around with deep learning a bit, I realized 2 things:

1. Yes, our phones can do this, but software like this is the  of technology that actually powers these effects in our phones. It’s like learning to program 2 + 2 = 4. A calculator can already do that, but it’s the first step in learning to program.
2. While having a computers create art in the style of famous artists may not change the world, that same technology WILL do and ALREADY IS doing incredible things. Powering translation software, self-driving cars, computers that detect cancer, etc. Deep learning powers all of it.

That’s what excites me so much about this!

Going from a computer recreating pictures in some artist’s style, all on its own, to world changing applications is not as huge a leap as it may initially seem.

Plus, seeing Grumpy Cat in the style of Van Gogh ain’t so bad.

What’s Next?

While I got the chance to play around with deep learning here a bit, I only understand the surface of it.

I can roughly explain the idea of what deep learning does, but can’t begin to explain how it works. That’s why there isn’t any technical information in this post.

But expect some technical info soon. Earlier today I spent some time building my very first neural network from scratch.

So keep an eye out for even cooler stuff coming soon!