deepart.ai

• 

  The icon of deepart.ai

Inspiration

Our vision is that users can generate abstract art which reassembles their taste of art. The main part of this project are evolutionary algorithms which evolve by getting the rating of the user for the generated pieces of art.

This project shows the potential of deep learning and evolutionary algorithms in the context of modern art and society. We want to fascinate users and show them, how far modern computing has gotten.

What it does

deepart.ai generates images based on polygons which are created by evolutionary algorithms and lets users rate the generated image.

How we built it

Our technology stack includes Python, lit-html (a subproject from Polymer) and Django (django-rest-framework). Each part of the project is represented as a microservice and will be orchestrated with docker.

Challenges we ran into

A huge part of the challenges we ran into were related to the generative and evolutionary algorithms. We needed to weigh the risk of overfitting the algorithms or making it too random/abstract.

By far the biggest challenge was implementing our own genetic and evolutionary algorithm. We needed to do this, because our use case is very special, so there was no existing library which could fit our needs.

Another challenge was connecting the vision and the backend components, because of the nature of APIs.

Accomplishments that we're proud of

We are proud of the communication of the backend and the vision component. We struggled a lot but we finally accomplished the communication using a message broker.

What we learned

We learned that computers can be used for art and to make beautiful and abstract content.

What's next for deepart.ai

We are planning on implementing a proper social network for deepart.ai so that users can share their generated art. We also want to create community networks so that multiple people can work on refining a network together to create beautiful, interconnected and abstract pieces of art.

Built With

• javascript
• keras
• lit-html
• python
• rabbitmq
• tensorflow