asingh33/CNNGestureRecognizer

Implements real-time hand gesture classification using a 12-layer CNN with dual preprocessing pipelines (binary thresholding and HSV-based skin masking) to handle varying lighting and background conditions. Features training capabilities to add custom gestures, in-app feature map visualization for network interpretability, and real-time prediction output to JSON or console. Supports both TensorFlow and Theano backends via Keras, with pretrained weights for five gestures (OK, PEACE, STOP, PUNCH, NOTHING).

1,043 stars. No commits in the last 6 months.