CNN model to recognize hand-written digits
This is a Convolutional Neural Network model to recognize hand-written single digits.
The model is built using TensorFlow Keras APIs.
This is the design of the model, using visualkeras:
The core of the model are three blocks of convolutional 2D layers with same padding, interspersed by max pooling layers and dropout (at 30%). After the convolutional blocks there is flatten layer which connects a feed-forward neural network with 50% dropout. The final layer is a softmax for classification.
Here the network performances:
The project includes a streamlit we-application to use the CNN model