We think raising public concern for climate change is extremely important and neural style transfer on images is a highly relatable way to achieve this.
What it does
It takes a user's image and repaints it in the style of a climatic natural disaster, using style transfer.
How we built it
We set up a Google cloud server and a React.js app. We set up an API for the app to communicate with the server. For the style transfer part, we modified an existing approach based on feature extraction from the VGG network.
Challenges we ran into
- Setting up the Google cloud server and enabling GPU support.
- Compatibility issues with our style transfer optimisation script.
Accomplishments that we're proud of
- Setting up the flask server and the scripts for communicating with it.
- Our awesome app and animations.
What we learned
Setting up VMs servers etc can be surprisingly time-consuming.
What's next for Climate gradient
Additional features such as location-based image feeds, more natural disaster styles and improved style transfer with a feedforward neural net.