Mistral

• 

  Drawing View

• 

  Style transfer in and of itself

• 

  Exploration View

• 

  Taking a photo for style transfer

• 

  Stencil Mode

Inspiration

We were inspired to create an app that promotes creativity and relaxation.

What it does

It provides a base for users to create artwork with the help of ai which can assist them in their creative endeavors.

How we built it

We use a pre-trained CNN for feature extraction of both images. Then it separates the important features from image that's gonna be converted. It defines content loss to match content features and style loss to match style statistics. We use a gradient descent model to minimize the total loss. Finally it iteratively adjusts the image till we have the final product.

Challenges we ran into

Using pytorch to create the model since none of us had sufficient experience. We also had to do the AI from scratch since we could not find any models online. Furthermore, there was a long compilation time.

Accomplishments that we're proud of

The app turned out fine and functions the way we intended. We are proud of the fact that the style transfer applied successfully and consistently. We are also proud of the fact that we trained the AI from scratch.

What we learned

We improved our programming skills, specifically with pytorch to create the model. We also learned how to use firebase storage, neural style transformations, and how to use swiftui.

What's next for Mistral

We will make necessary improvements to make the app more mainstream and accessible. We plan to add more features such as an adaptive AI learning mode.

Built With

• firebase
• python
• pytorch
• swifui
• uikit