Alpha(bet)Recognizer is convoluted neural network(CNN) trained on hand drawn images of the English alphabet from the EMNIST dataset.
AlphaRecgonizer is unique because it leverages a deep learning technique called Transfer Learning, training a model ontop of a pre-existing model.
AlphaRecognizer is trained ontop of the RESNet-50 model for 15 epochs. After 15 epochs, it is then trained with its base layers unfrozen for 4 epochs.
AlphaRecognizer has an accuracy of 99% but it still needs some help learning. So draw some letters to confuse the model and help it learn!
AR Canvas
To get started draw any alphabet in the English language in uppercase in the black canvas below.