Thomas was driving one day, and he heard a talk on the radio that was a debate on how people view beauty. They claimed that, beauty can't be put on a scientific level, and it was based on the bias of people, which differs from person to person. Hearing that, Thomas wanted to build an a.i. that can quantify and try to generalize how one person can view another person's beauty compared to their colleague. With the help of Tommy and Sasha, our team was able to implement an neural net. that, with the help of picture training, could try to scientifically quantify a person's theoretical beauty.

What it does

It takes in a picture, and uses the neural net to detect and map out their face with 68 points. Then, it goes through the back-end, and through the personalized neural network, it estimates the person's theoretical beauty in your eyes.

How I built it

We used Android studio as well as fire-base to secure, and process data. We also used gicheonkang's fast_face demo as a basis for our face detection and then created everything else in the app including the UI, the back-end, the user authentication, the neural network and we also had modify the face detection algorithm in order to meet our requirements.

Challenges I ran into

Lots of bugs, and the fact that with the pictures we were feeding it, lots of the references were further away than expected. So, that means, the same person who was closer, would result in different data sets compared to the same person, but standing further back in the picture. We also ran into a lot of fire-base data issue, where data being passed was not passing correctly.

Accomplishments that I'm proud of

We were able to make a small neural net that can see faces and collects the points and trains it to scientifically give out a quantifiable result.

What I learned

How it is a pain to work with android studio if the versions downloaded were different. We ran into a lot of gradle sync issues. Also, that working with training data is hard, if the data being uploaded is not being input correctly, which means, more time training the a.i.

What's next for NeuralFace

Expand the u.i to be better, and maybe expanding it to other platform. Please hire us.

Built With

Share this project: