Deep Learning applied to Deep Thoughts

Thanks to my host who invited me along, I was visiting the Museum of Modern Arts in Manhattan New York City last week. The most famous item of the Museum collections is a painting by Vincent van Gogh, called The Starry Night.

The painting was hanging on the wall since a donation to the museum in 1941 ... don't ask me how much it worths, must be millions, if not billions. Anyways, it is unique in the world and a masterpiece.

While I was trying to take a photo, the safety guard reminded me that photographers must be at least two meters away from the painting... Suddenly, it reminded me of "TF Stylize" app in my pocket, which can convert a smartphone's camera into an artist's paintbrush. The Android app is one of the showcase examples from Google's Tensorflow, an open-source deep learning library, where you can build from scratch here: https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/android.

Once the app fired up, there is a panel of 26 existing paintings to choose from various artistic styles. Guess what? The Starry Night is one of them. Ah ha! It was not a coincidence that I came across this idea impromptu.

Now I had the luxury to point the Stylize app towards the painting frame. It would stylize the painting using the style of the original painting itself, just in front of my naked eyes.

Now you see this unique picture as a result:

Tell me what you think of the picture. Is it more abstract, more beautiful, more impressive, or more absurd? Thousands of readers may have thousands of Hamlet, you will have your own judgment. My job as a "Time Traveller" who brought you back to the minds of the Master Van Gogh is done here. Maybe in his own eyes, this painting would not worth that much of appreciation. It must be the people in the asylum who needed to treat him better while he was alive ...

To be honest, you may reproduce the picture next time holding the phone like this at the museum, but chances are you may get stuck in a long queue and do not have enough time to download the app to do the shot :-)

Now it comes to the hard bit: if you would like to know more about Deep Learning, I am going to write more articles to explain the concepts using concrete examples like this. Stay tuned !