Sign Language Recognition (ASL alphabet)

A small CNN that recognizes hand signs for 24 letters of the American Sign Language alphabet (A-Y, excluding J and Z — those require motion).

The model was trained on Sign Language MNIST — 28×28 grayscale crops with uniform background and hand centered. To keep predictions honest to that domain, this demo runs classification on per-letter samples from the test set rather than arbitrary photos. Pick one below.

Source on GitHub

Click a letter sample (one per letter, from the test set)
Pages:

Pick a sample below to classify it.

(J and Z aren't supported — they require movement.)