ASL-to-text

This project is a submission to HackHarvard 2022.

DEVPOST

Models

Models are trained manually using a live 3D point cloud of your hands using mediapipe. We campture multiple points per second to achieve a large enough dataset to train our chosen model structure, PointNet.

abc model

Trained model that recognizes every latter in the English alphabet (using Signs).

conversation model

Trained model on the basics of ASL.

Words used:

• yes
• no
• hello
• bye
• how
• old
• you
• me
• explore
• deaf
• good
• bad