AAST- Andrew's Art Style Transfer

Inspiration

Art Style Transfer is a hot topic in recent years, accompanied by the prosperity of AI. Generally, two images are needed. One is the content image, and one is the style reference image. The output images should be a blend of the two images.

What it does

This project can analyze the features and transfer the art style from one to another. People can have fun with it. It can be applied to art creation and entertainment.

How I built it

A popular method is deploying CNN (Convolution Neural Network). And that's what the operator wants to use in this project. I referred to the Tensorflow GitHub Repos and read an article about art style transfer. Their links are in the project README.md

Challenges I ran into

1. How to deal with different backend framework (TensorFlow and Keras)
2. How to reduce the dimension of images and analyze their features

Accomplishments that I'm proud of

The project is able to generate a blended artwork within one hour, and the score(feature loss) is decent.

What I learned

1. Gradient descent
2. Adam Optimization
3. VGG19 library of Keras

What's next for AAST- Andrew's Art Style Transfer

A mobile app and a website to blend art styles.

Built With

• jupyter
• python