DeepArt.io is pushing the boundaries of machine learning in the cloud, in a really beautiful way.
https://DeepArt.io/
DeepArt.io allows us to upload a photo of something or someone, then we are instructed to upload another image that will serve as a style. The machine takes both images into account, creating something totally new and unique, all on it's own, with no further human intervention.
It feels like we are allowing the computer run wild with PhotoShop doing whatever it thinks we want, and it seems to be pretty good at it too!
DeepArt.io Examples
We start by out adding a simple image of a lion...
Then a style image was applied to it, that looks like this...
After taking some time to think about it, the machine generates the following original work of art...
Pretty amazing right? There are many more community generated images for your enjoyment at the following link...
How Does This Work?
The details on the dark magic they do will always be clouded to laymen like us. In terms we can understand... The computer is producing art in a way that simulates human thought.
DeepArt specifies that their system uses a neural network to identify the subject matter of each image, so that they can be re-combined using the complicated interplay between content and style. In the end, the computer has created a custom piece of art. Which is something that humans do regularly.
For example, we take a photo with our wife on our wedding day.
Then later on, we see an image of stained glass.
In our heads, we are able to identify the wedding moment and the stained glass style, in order to re-create the image in stained glass.
I am not a computer scientist or mathematician so please don't take my word for it. That is just my best attempt at understanding their technical explanation in the Cornell University Library.
Fascinating stuff!
My Own Examples
I signed up to DeepArt.io and started uploading images.
I pulled this image of Jimi off of Google...
Then I pulled this image of smoke off of Google to act as the style...
Next, I instructed the DeepArt service to apply the smoke as a style to Jimi, and it did so in a rather wonderful way! Look at the the jacket he is wearing and how it used smoke to draw the decorations. Beautiful, and strangely similar to what my brain imagines if told to do the same thing.
Let's Do Another One!
This time I used all my own original photography by starting out with this photo from my introduceyourself post.
Then I applied my photo from Trump Tower Chicago at night.
The DeepArt service returned the following image!
Try Another Style!
Oh this is fun! I used the same image of myself...
Then as a style, I applied a comic book character that I drew.
After thinking about it, the machine realized I wanted a comic book version of myself...
No More Secrets
I am not going to give you the ingredients of the next images, but I had them created by the DeepArt.io machine!
Video Application
My mind is going nuts thinking about the application here. Imagine if they were not still images but video! Like this!
The world of digital video and movie effects processing could totally benefit from machine learning. There is probably a huge world of wonder awaiting once we really start to turn the machine loose. The future looks groovy!
