Bana:
https://github.com/banaloul-cy/ASL-Live

Bana:
ASL Live is a Python desktop application that helps users explore American Sign Language fingerspelling through a live webcam preview and an alphabet reference guide.

The application uses MediaPipe to locate the hand and a pretrained Vision Transformer model by dima806 on Hugging Face to classify hand images. A geometric finger-landmark check helps distinguish L and Y. The English interface features an olive-green theme, a large detected-letter display, and camera controls.

Our work includes the desktop interface, webcam integration, hand cropping, prediction filtering, and the L/Y hand-shape check. We use existing pretrained model weights rather than training a model from scratch.

The current prototype focuses on static alphabet signs. It does not translate full sign-language sentences or recognize the movements required for J and Z. Predictions may be incorrect, and accuracy has not been formally evaluated.

After the initial model download, webcam inference runs locally on the user’s computer.