This app starts empty — you must define and capture every gesture you want the model to recognize.
Follow the steps below to add reliable gestures and get good recognition results.
How to add a new gesture
- Pose the new gesture clearly in front of the camera. Keep your hand centered and steady for a moment.
- Type a descriptive gesture name in the label box (for example: "thumbs_up", "wave_right").
- Press Capture to save the current frame. Capture multiple photos (5–20) from slightly different angles and distances for the same label.
- After collecting enough samples for each gesture, press Train Model and wait for training to finish.
- Press Start Recognition to test your gestures in real time.
Recommendation: keep gestures distinct
Create clearly different labels for gestures that look or move differently. If two shapes are similar, give them separate labels (e.g., "point_left" vs "point_right" or "open_palm" vs "flat_palm") so the model can learn each one separately.
Avoid using the same label for visually different poses — if a gesture has meaningful variations, capture each variation under its own label.
Practical tips:
- Use good lighting and a plain background when capturing photos.
- Capture both hands and slightly different rotations if the gesture may appear that way.
- Collect varied examples (angles, hand position, distance) for each label — more diverse samples improve accuracy.
- If recognition confuses two gestures, add more samples for each and consider renaming to emphasize differences.