Skating Skill Scoring

Video Uploading UI

Inspiration

Game-like style scoring to transform video submissions into a powerful, community-driven platform for tracking progress and elevating every skater's skill level.

What it does

Receives video files of skateboarding tricks and uses a computer vision model to grade and score the skateboarding trick. Then, it ranks it against other users in a global TAMU leaderboard.

How we built it

This application is built as a full-stack web service, separating the user interface from the logic. The frontend is a single HTML, Tailwind CSS, and JavaScript file that handles video upload and display, while FastAPI powers the backend and uses Supervision, OpenCV and YOLO libraries for python to manage the submission process, run the scoring logic, and persist all data in a SQLite database.

Challenges we ran into

Training our Board Model: Training one of our vision models to detect where the skateboard is took upwards of 6-ish hours and halted multiple times due to bad datasets.

Hooking everything together: Actually making all of our separate pieces of logic talk to and interact with each other was probably our #1 issue.

Accomplishments that we're proud of

Our final board model accurately detects the board with high precision after only training for 50 epochs with ~300 images total.

What we learned

We learned how to effectively communicate as a team and resolve issues with our code base. On the technical side, we learned a host of new python libraries for computer vision and how to effectively find datasets. We also learned a lot about Javascript, FastAPI, and HTML on the front end.