Artify

Inspiration

A new algorithm was developed recently which takes the content of one image and the style of another to make a new image. I thought it would be cool to make an easy-to-use app incorporating the algorithm.

What it does

Using deep convolutional neural networks, the content image is gradually changed into a new image with similar activations to the original and similar correlations in activations to the style image. The user currently supplies the content image, and selects from some provided style images.

How I built it

The images are produced using Anders Boesen Lindbo Larsen's implementation of the algorithm, running on a server with heavy GPU processing power. The user interacts with a web interface built with jQuery and Semantic-UI, supported by Flask, and is able to view the transformation in real time.

Challenges I ran into

The algorithm is quite demanding, so we were only able to use a server with GPU processing available for the app to be usable in any sense of the word.

Accomplishments that I'm proud of

We were able to put it together in less than 24 hours.

What I learned

For Jun, it was a great way to dive in to web development for the first time. For Daniel, it was a great way to try out a new web framework.

What's next for Artify

Deployment on an actual server, rather than Daniel's desktop, would be great. However, the algorithm is still quite demanding, so improvements in the efficiency of the algorithm itself would be necessary to move forward to deployment.