Rock Paper Scissors CV

Installation

Front End

• The front end was created utilizing Angular v18. This application will require access to a webcam/camera.
• Run npm install to install dependencies. To run the front end, run ng serve or npm start.

Back End

• The back end utilizes FastApi and Uvicorn running with OpenCV, GestureRecognizer, and Keras. Run pip install -r requirements.txt to install all dependencies. Then, run main.py to run the API client.
• Note: Inside the backend folder exists a notebook folder with two notebooks that are not used in the API calls/model prediction.
  • main.ipynb utilizes the OpenCV capture and the hand prediction can be done locally.
  • crop.ipynb uses GestureRecognizer to crop out hands from a dataset, which was used to train the model. This code is not part of anything running in the model prediction.

Instructions

• Raise a hand up and choose a paper, scissors, or rock form with it. Click Take Picture and face against the computer in a simple game of Rock Paper Scissors!

Model Training Notes

• This model was trained utilizing a CNN for image classification utilizing a variety of Rock Paper Scissors Datasets. The Kaggle/Jupyter Notebook can be found here.
• Overall, the model was trained several times and gained 92.73% accuracy on the test data: