Trained a CNN model on the MNIST dataset.
-Used 2 convolutional layers of window size 5x5 and 3x3 and added 2 dense layers in the end.
-Data Augmentation to improve accuracy of model on user inputted data.
-Used Adam optimiser to achieve an accuracy of 98.9%.
-Saved the Model in tensorflow js format and loaded in into the js file.
Used Canvas HTML API for the user to draw the digits. Added JS Event listeners.
Works on both touch interfaces and computers.
Predicts the Digit drawn by the user from 0-9 with confidence level:
Check out live website at: https://sarthak17jain.github.io/handwritten_digit_pred/
