Dr.ProPic

Inspiration

One of our housemates was having trouble making his tinder profile, so we decided to take a tech approach to solve the problem of choosing which pictures to use, by training a machine-learning algorithm to do the job for us.

What it does

Our marketing team gathered research on what a wide sample of people considered to be attractive, and we hand trained this data in house by having that wide sample be our teammates. The algorithm will accept a zip file of pictures and will rank those pictures from 1-10, and output a text file for the user to view the results

How we built it

Using Microsoft Azure's cloud-computing based CustomVision API, paired with a dataset that was hand trained by our teammates, we were able to produce a machine learning model that can accurately rank how attractive somebody is.

Challenges we ran into

There were very few datasets readily available labeled with attraction data, so we had to spend a lot of time deciding which labels to use for our model, and getting an accurate model required a substantial amount of training data, so that was incredibly time-consuming. Another challenge was that Javascript was having trouble accessing a server from Azure so routing any Python scripts was impossible to test for me, which forced us to switch from a website to a Python-based GUI. I didn't have enough time to learn any other kinds of GUI's, and TKinter was incredibly easy to quickly set up.

Accomplishments that we're proud of

Learning Azure to the extent of having the ability to be able to produce, and actively use Machine-Learning models, was an incredible achievement, as well as trying to learn HTML, CSS, and JavaScript. We also got direct experience using API's to do significantly more powerful things than we would've been able to do on our own, which was a cool experience in and of itself.