This project is a MoveNet-based pose estimation demo built with React and Node.js. It uses TensorFlow.js for in-browser pose detection and provides a modular, interactive frontend.
- Real-time pose estimation using MoveNet (MultiPose)
- Modular React frontend
- Offline model loading (no internet required at runtime)
- Multiple interactive demo pages
- Stick Figure Disco
- Simon Says : Do a pose and get a reward
- Short Term Memory : Find the previously shown object
- Long Term Memory : Find the previously shown object
- Evade the Bear : Identify aversive auditory stimulus and move away
- Evade the Owl : Identify aversive visual stimulus and move away
- Quz : Answer Questions with options by making a certain posture
More interactive demonstrations are welcome!
- Node.js (v16 or newer recommended)
- npm (comes with Node.js)
- Clone this repository or download the source code.
- Open a terminal in the project directory.
- Install dependencies:
npm install
Start the development server:
npm run dev- The app will be available at
http://localhost:3000(or the port shown in your terminal). - Make sure your webcam is connected and allowed in your browser.
npm run dev— Start the development servernpm run build— Build the app for productionnpm start— Start the production server (after build)
This project was created with the assistance of GitHub Copilot and GPT-4.1.
This demo was made by Peter Christ and Aditya Iyer, for the MPINB Bonn Open Day. Thanks to Laurin Büld, Eric Jelli, Anja Günther for assistance with debugging and setting up the demonmstration.
Images for this demo were either generated with Gemini, or obtained from flaticons (Freepik, NajumNahar, pocike)
Feel free to use with attribution.