Poser

Inspiration

My teammates and I wanted to tackle an issue that would be challenging tech-wise. We were inspired by the dance dance revolution arcade machines at Price Center but we didn't wanna travel from ERC to use the machine. This lets us access a pose-off from our fingertips using basic laptop hardware.

What it does

Poser projects a livestream of the user, playing notes in a sequential order as the user poses correctly to the queue given.

How we built it

We utilized React in order to build a responsive app. In order to add Tensorflow functionality, Tensorflow.js was planned to be used after repurposing a pose estimation computer vision model through transfer learning on AWS. We got data for transfer learning by taking many pictures of our teammate Akhil posing.

Challenges we ran into

We had issues finding properly documented machine learning models for use in transfer learning. Ultimately we compromised on using a MobileNetV2 architecture built for ImageNet rather than pose estimation, readjusting the weights using our data.

Accomplishments that we're proud of

We're proud to have built a React app - many challenges were faced, but in end each challenge helped promote our understanding of React. We also enjoyed the opportunity to go through a transfer learning process for a practical application.