This project is part of the programming lab for the UMN CSCI 8980 course: Visualization with AI.
In this project, an ML model will predict directions from an image in web camera. We can fine-tune a pretrained MobileNet model to predict 4 different classes (i.e, up, down, left, right) as defined by the user.
Check out the Live Demo
- Add Example: Use your web camera to provide example images for the four different classes (up, down, left, right).
- Train: Fine-tune the ML model with these images.
- Play: Start playing Pac-Man by making gestures in front of your web camera.
- Ensure you have a code IDE like VSCode installed. VSCode is recommended, but feel free to use any IDE of your choice.
- Install npm on your machine.
- (If you are familar with Github)
fork
this repository, andclone
the forked repository to your local machine. Learn how to fork and clone. - (If you are not familar with Github) directly download the code from this repo
- Open the cloned project folder with VSCode.
- Launch the VSCode integrated terminal from menu:
View > Terminal
. - In the terminal, run
npm install
to install necessary npm packages (first-time setup only).
Execute npm start
in the terminal.
The project will open in your default web browser. Any code changes will automatically update the webpage. Google Chrome is recommended for the development.
This project is a modified version based on the official TensorFlow.js demos. For more information and additional context, visit TensorFlow.js.
The PANC-MAN uses the implementation at https://github.com/astraube/pacman-covid-19/