For 2022, I decided to build a system that helps me take a picture of myself every day. The goal is to make a time lapse of how I change over a single orbit of the earth. I use facial feature detection to find my eyes and then instruct myself how to move to get into the same position for ever picture.

This Instagram post shows how the facial tracing works:

https://www.instagram.com/tv/CYNXVkzKs4X/?utm_medium=copy_link

But then I had the idea:

What if I train a variational auto encoder on the pictures too. So using TensorFlow, I built a VAE using a number of 2D convolutional layers as well as a few others to compress the data into a 16-length latent vector.

Once it was trained l, I built a webapp that allows you to explore the latent space.

Try it out yourself

I’ll retrain the VAE at the end of every month of pictures.