Corgify

cat
corgification

Inspiration

We saw that image translation is possible using Generative Adversarial neural networks. In particular, we we love this example where you can transform a sketch of a cat using this edges-to-cats model.

edgetocat

We wanted to try transforming any dog breed to corgi. Finally you can sleep at night, knowing what your inferior poodle would look like if she were a GLORIOUS corgi.

corgification

What it does

We wanted to train a generative adversarial neural network to be able to generate a convincing looking corgi. Given an input of a random dog, the network would ascend that dog to corgi status.

How we built it

We made use a Google Cloud Services virtual machine for deep learning. We made use of the Stanford dogs dataset for training the network. Since we're neural network noobs, we followed a TensorFlow Pix2Pix tutorial to try to hack everything together. We did all the programming in Python, in a Jupyter notebook.

Challenges we ran into

The only thing we knew going into this challenge was Python. We knew we were being ambitious, but in the hackathon spirit, we had to give it a shot!

Firstly, we'd never used Google Cloud Services before. It took us several hours to get it set up, and several more hours to iron out some problems we had with e.g. permissions.
Secondly, we'd never made a neural network before. It took a while to get the hang of everything.
We got stuck on making the data pipeline to feed into the neural network.

Accomplishments that we're proud of

Picking an ambitious project and doing our absolute best even though it was beyond us.
The vast amount we learned.
Actually managing to work with Google Cloud.
It's Khafarae's first ever hackathon.
Pushing through and overcoming a lot of problems.