Simple real time digit recognition application developed in python, where user draws a digit in the GUI and application predicts the drawn digit.
- Training data and test data are loaded using mnist dataset provided by keras. Then the data is reshaped to the shape which is our model expects while training.
- CNN sequential model is created and compiled.
- The model is trained on the training data and the trained weights is stored in a file named 'mnist.h5'
- An interactive window is created using python Tkinter library.
- The testing image should have digit in white and rest all background as black.
- Load the pretrained model weights and use it on image to predict the digit.