Make my emoji

• 

  final output of the project

Inspiration

I was really intrigued by how Generative Adversarial Networks work and was looking for a quick application of them to work on. This app takes benefits of GANs final output in which Generator eventually becomes so good that it actually is able to fool a discriminator, and when this happens we have our output.

What it does

It is very similar to bitmoji in which you create your own emoji which looks similar to you face. But instead of customising each feature of face (jaw, eye brows, hair etc.) step by step you just upload your photo and it creates it for you.

How I built it

I used Tensor flow to build the Generative Adversarial Network in python.

Challenges I ran into

Time limitation, I don't have a GPU in my machine and it takes a lot of time to train.

Accomplishments that I'm proud of

I understood working of GANs which I wanted to do. This is a kind of style transfer project that we have seen in many apps these days which makes your portrait looks like it was made by Picasso or DaVinci. I wanted to this project from so long and used this opportunity to work on it.

What I learned

Learned the basics of GANs, read the Ian Goodfellow paper. The work I've done in this project is actually a reproduction of a paper published by Facebook's AI Lab. Reading and understanding both these papers helped me gain a lot of knowledge about math behind machine learning. The code was written in Tensor Flow which helped me strengthen my basics on tensor flow.

What's next for Make my emoji

This is currently just a model in my system, I would like to export this model to my iPhone using Apple's ML toolkit or create a web service of it. Using this I would like to make an application or website in which users clicks the image and sends it to the service and receives back his emoji image.

Built With

• python