The Geography of Home
Category: Machine Learning
Team: Myself
Role: Entire project
Date: December 2016
Background
This is my final project for my Machine Learning class at NYU ITP taught by Patrick Hebron. We were given the freedom to design anything as long as we used machine learning.
Design idea
These are some map screenshots of the places where I lived in Romania, Germany & US. In the far right in the upper and bottom screenshots as you can see the roof of the houses have a red brownish color. That color somehow feels like ‘home’ to me. These two screenshots are from my street in Romania, where I grew up. If you look closely, the other colors of the other places where I lived are very grey-is and the rooftops are all grey.
Discovery
These are two pictures of my hometown Cluj - Napoca in Romania and you can see the colors of the roof being again red-brownish. So I came up with this idea for my project to perform a style-transfer on the screenshots that had grey rooftops using these 2 pictures.
Training
Training took about 88 hours. Some pictures went through 3,550 iterations, while others only through 400 iterations.
Results
Here are the results. In the first picture, I performed a style-transfer with one input of the current place that I lived back then in NYC in order to make this place feel more like my hometown in Romania. The output you can see on the right side. In the picture below, I performed a style transfer with 2 inputs and the output you can see on the right side. The output that used 2 inputs looked more clear and went through a lot more iterations.
The Geography of Home, Re-imagined
And this is how the geography of home, re-imagined looks like. These colors definitely feel like a ‘warm’ and cozy place for me and closer to home. This project was an exploration data visualization where I incorporated machine learning to learn more about the infinite ways of how AI and machine learning can improve our lives and be incorporated in any projects.
