A Flask web app for handwritten digit recognition using a CNN architecture. This model is trained on the dataset from Kaggle in Tensorflow. Got an accuracy of 0.99614 in this contest.
CNN Architecture:
[32C3-BN]*2-[32C5S2-BN]-0.4d-[64C3-BN]*2-[64C5S2-BN]-0.4d-F-256D-10D
nCm -> n CNNs with size mXm
BN -> Batch Normalization
xd -> Dropout of x
F -> Flatten
xD -> Dense layer of x units
Epochs : 20