FINAL CHALLENGE / PRETRAINED VISION → PLAY
Show your hand.
Rock, paper, scissors—with your camera as the controller. No examples to collect. No model to train.
Camera on. Game on.
Allow the camera, start a round, then show one clear hand when the countdown says SHOW. Hold it steady until the gesture locks.
Choose camera or the illustrated practice mode.
Camera permissions · Windows and Mac
Windows: Settings → Privacy & security → Camera. Allow your browser and this site to access the camera.
Mac: System Settings → Privacy & Security → Camera. Allow your browser, then allow this site’s camera prompt.
Choose an input, then start a round.
Your hand
Waiting for a hand.
Practice mode: click the hand you want to play after SHOW.
Make your move.
The computer locks its move before your gesture is read. Can you win?
First to five? Play a few rounds and see.
Recognize, then respond.
A pretrained Google MediaPipe Gesture Recognizer supplies hand landmarks and gesture labels. We map Closed_Fist → rock, Open_Palm → paper and Victory → scissors. A stable label locks your move; game rules choose the computer’s response.
Classic: the computer commits before SHOW. Counter move: it sees your recognized hand, then deliberately chooses the winning response. That mode demonstrates recognition and a rule—not early prediction.
Camera frames are processed locally. No video is saved or uploaded. Practice buttons demonstrate game behavior without testing camera recognition. The original lecture’s tiny CNN explains training; this game uses an already-trained model.
See the hand model’s structure and credits →