Disco Train

Inspiration

This project was inspired by our love for dance.

What it does

This program has a front-end landing page demo for a website for a dance-sharing platform, and back end support for scoring videos based on the similarity of the dance moves. Given two videos, we identify key points in the user's movement and compare them between the videos to create a score describing how closely they are related.

How we built it

Built with HTML, CSS, JS, Python, and SuperGradient.

Accomplishments that we're proud of

The front-end program is sophisticated and shows many avenues for future expansion. The back end program can identify joints in the image and generate a sequence of key point locations, which are saved to a file for later use. These sequences allow for real time comparisons between two images and an automatically generated score to describe the relationship between the two images.

We developed two different concise algorithms that allow for accurate detection of the human body to emulate dance moves, and allow for comparisons between the inaccuracies of said models

What we learned

We learned SuperGradient, as well as advanced our skills in python, HTML, CSS, and JavaScript.

What's next for Disco Train

Disco Train's website is design with expansion in mind. The nav bar displays some of the functionalities that we feel would enhance the project, such as pages to join live dance battles, dance on your own, and create new dances. Linking the website up with a database to store user information and dance videos would allow users to browse videos other users have made and save their own videos, as well as implement the search bar properly. The model could be further optimized to use CUDA to allow for faster computations for the key point localization.