Inspiration
In addition to feeling remarkably disconnected to other people, a lot of people are struggling to gain the motivation to get up and start moving about during the pandemic. {T}w{e}rk Out! aims to solve all of the above - a fun dancing game that gets you off your bum and burn calories with your friends.
What it does
You join a room with your friends and pick a song to dance to. Then you will try a repetitive dance move (such as the "floss") for a certain amount of time and whoever does it most (a.k.a. burns the most calories) will get the most points!
How we built it
Training the model
- We automatically captured movements for popular dances (such as flossing, can can, disco) using PoseNet using ML5, react, and P5.js
- After collecting datapoints for the poses, we train an SKLearn model to classify the stages of the dances
Video feed
- We used Twilio to allow two people to communicate with each other through their webcam and pass the frames to the ML model
Prediction
- We detected the poses from the live video feed and classified them using our model
- More accurate movements entail a higher score
- The scores are stored on Firebase and are displayed live on the React front end
Challenges we ran into
Third party libraries are really bad. You can only trust yourself (and SKLearn)!
Accomplishments that we're proud of
It works
What we learned
Don't do react in hackathons
What's next for {T}w{e}rk Out!
More dance moves trained on real dance experts
Built With
- firebase
- google-cloud
- ml5
- p5
- posenet
- python
- react
- twilio
Log in or sign up for Devpost to join the conversation.