Deep Shirt

Inspiration

• Bored of the usual “Off-the-shelf” designs, we wanted to enable users to create unique T-Shirts with custom prints and unique style.
• Empower the user to express his creativity every day.

What it does

• Our application generates personalized T-Shirt styles with the help of Machine Learning and the Cloud. The user picks the content image and a particular style image. Both images are merged into a unique style, which is then printed on a T-Shirt.

How we built it

• Our application is separated into two main components: The Frontend and a Compute Instance for machine learning.
• The frontend is dynamic web application built with React and Twitter Bootstrap to enable an interactive user interface. For the deployment of this component we use Docker.
  • In the machine learning component a Tensorflow model is running on a GPU instance from Google Compute Engine.
• As a data store and database we rely on Firebase.
• To create the individual T-Shirt we use the Mockup Generator from Printful.
• To support the image selection process we use the Flickr Api, such that the user can search images.

Challenges we ran into

• Setting up the whole pipeline, making it as intuitive as possible for the user.
• Setting up Google Cloud Engine to run Tensorflow models.
• Creating HTTPs certificates.

Accomplishments that we're proud of

• Built an entire web application from scratch in a constrained time.
• Distributed the architecture across multiple servers, integrating a powerful GPU Machine from Google Cloud.
• Provided a pleasant user experience for selecting images and seeing the results.

What we learned

• React.
• Tensorflow.
• Style Transfer.
• Firebase Storage and Authentication.

What's next for Deep Shirt

• Allow the users to purchase the newly created T-Shirts.
• Integrate a rating system to rank the best T-Shirts on the home page.

Repo

Built With

• css
• docker
• firebase
• go
• html
• javascript
• letsencrypt
• machine-learning
• python
• react
• redux