Inspiration
Our inspiration for this project comes from the pandemic. Traditionally, the art industry has always focused on the physical aspect of art. Physical paintings, museums, viewing art in person, etc. During the pandemic, this was very difficult and art interaction decreased because most museums and in-person viewing events were restricted. The COVID-19 pandemic has almost forced the art industry to look at digital art and other ways of creating and experiencing new art. So we decided to make our own digital art.
What it does
Our project takes in two images: a “content” image and a “style” image. The content image could be an image of anything, an animal, the Old Well, even Rameses. The style image is generally supposed to be a painting from a famous artist. Our project combines the two images by making the content image the foreground and the style image the blueprint for the image. Essentially, the algorithm will take the content image and adapt it to the style of the artist in the style image.
How we built it
We built the image art styler by utilizing a tensorflow hub module which is an arbitrary neural artistic stylization network, based on a paper presented at a machine vision conference. You can check out the paper at (https://arxiv.org/abs/1705.06830). This is essentially a specialized neural network that helped us efficiently combine the two images.
All of the code and back-end development was done in a Python Notebook in Google Collab.
We built the website using React for the front-end development and it is hosted on a team member's portfolio website. All code is stored in GitHub.
Challenges we ran into
One challenge we ran into was time. These challenges are further described in the future of the project section.
Accomplishments that we're proud of
We are very proud of the project that we have created. It looks great and we had a ton of fun making it. We didn’t even know any of this was possible at the beginning of the weekend and now we are able to generate some great images as a modern day revival of old artists’ work, but done by a computer!
What we learned
We had never implemented, and some of us even heard of, neural networks! It was cool being able to combine our knowledge and figure out how to do something really amazing with Python and React.
What's next for Image Art Styler
The images that we generated took only a few seconds to combine. To take Image Art Styler a step further, we could write code and create algorithms that maybe take a little longer to develop, but increase significantly in quality. For example, it was not feasible within the timeframe, however some others who have used similar tools have added loss functions and used gradient descent methods to make higher quality images over many iterations. In the future, we would have created loss functions to optimize our images, but for now we have a quick and easy method of combining the content and style.
Log in or sign up for Devpost to join the conversation.