Going deeper

Image Modeling with Keras

Ariel Rokem

Senior Data Scientist, University of Washington

Network with one convolutional layer

Network with one convolutional layer: implementation

model = Sequential()
model.add(Conv2D(10, kernel_size=2, activation='relu', 
                 input_shape=(img_rows, img_cols, 1)))
model.add(Flatten())
model.add(Dense(3, activation='softmax'))

Building a deeper network

Building a deep network

model = Sequential()
model.add(Conv2D(10, kernel_size=2, activation='relu', 
                 input_shape=(img_rows, img_cols, 1), 
                 padding='equal'))

# Second convolutional layer
model.add(Conv2D(10, kernel_size=2, activation='relu')

model.add(Flatten())
model.add(Dense(3, activation='softmax'))

Why do we want deep networks?

Features in early layers

Features in intermediate layers

Features in late layers

How deep?

Depth comes at a computational cost
May require more data

Let's practice!

Image Modeling with Keras

Preparing Video For Download...