191 lines
5.6 KiB
Python
191 lines
5.6 KiB
Python
"""Residual blocks: Conv and identity"""
|
|
|
|
from keras.initializers import glorot_uniform
|
|
from keras.layers import Add, Activation, Conv2D, BatchNormalization
|
|
|
|
|
|
def basic_block(X, f, filters, s, label):
|
|
"""
|
|
Residual net identity block.
|
|
|
|
Arguments:
|
|
X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)
|
|
f -- integer, specifying the shape of the middle CONV's window for the main path
|
|
filters -- python list of integers, defining the number of filters in the CONV layers of the main path
|
|
label -- label to use for naming
|
|
s -- strides
|
|
|
|
Returns:
|
|
X -- output of the identity block, tensor of shape (n_H, n_W, n_C)
|
|
"""
|
|
F1, F2 = filters
|
|
X_shortcut = X
|
|
|
|
# First component of main path
|
|
X = Conv2D(
|
|
F1,
|
|
kernel_size=(f, f),
|
|
strides=(s, s),
|
|
padding='same',
|
|
name='conv_%s_2a' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2a' % label)(X)
|
|
X = Activation('relu')(X)
|
|
|
|
# Second component of main path
|
|
X = Conv2D(
|
|
F2,
|
|
kernel_size=(f, f),
|
|
strides=(1, 1),
|
|
padding='same',
|
|
name='conv_%s_2b' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2b' % label)(X)
|
|
|
|
# Shortcut path
|
|
X_shortcut = Conv2D(
|
|
F2,
|
|
kernel_size=(1, 1),
|
|
strides=(s, s),
|
|
padding='valid',
|
|
name='conv_%s_1' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X_shortcut)
|
|
X_shortcut = BatchNormalization(axis=3, name='bn_%s_1' % label)(X_shortcut)
|
|
|
|
# Add shortcut value to main path, and pass it through a RELU activation
|
|
X = Add()([X_shortcut, X])
|
|
X = Activation('relu')(X)
|
|
|
|
return X
|
|
|
|
|
|
def identity_block(X, f, filters, label):
|
|
"""
|
|
Residual net identity block.
|
|
|
|
Arguments:
|
|
X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)
|
|
f -- integer, specifying the shape of the middle CONV's window for the main path
|
|
filters -- python list of integers, defining the number of filters in the CONV layers of the main path
|
|
label -- label to use for naming
|
|
|
|
Returns:
|
|
X -- output of the identity block, tensor of shape (n_H, n_W, n_C)
|
|
"""
|
|
F1, F2, F3 = filters
|
|
X_shortcut = X
|
|
|
|
# First component of main path
|
|
X = Conv2D(
|
|
F1,
|
|
kernel_size=(1, 1),
|
|
strides=(1,1),
|
|
padding='valid',
|
|
name='conv_%s_2a' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2a' % label)(X)
|
|
X = Activation('relu')(X)
|
|
|
|
# Second component of main path
|
|
X = Conv2D(
|
|
F2,
|
|
kernel_size=(f, f),
|
|
strides=(1, 1),
|
|
padding='same',
|
|
name='conv_%s_2b' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2b' % label)(X)
|
|
X = Activation('relu')(X)
|
|
|
|
# Third component of main path
|
|
X = Conv2D(
|
|
F3,
|
|
kernel_size=(1, 1),
|
|
strides=(1, 1),
|
|
padding='valid',
|
|
name='conv_%s_2c' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2c' % label)(X)
|
|
|
|
# Add shortcut value to main path, and pass it through a RELU activation
|
|
X = Add()([X_shortcut, X])
|
|
X = Activation('relu')(X)
|
|
|
|
return X
|
|
|
|
|
|
def convolutional_block(X, f, filters, label, s=2):
|
|
"""
|
|
Residual net convolutional block.
|
|
|
|
Arguments:
|
|
X -- input tensor of shape (m, n_H_prev, n_W_prev, n_C_prev)
|
|
f -- integer, specifying the shape of the middle CONV's window for the main path
|
|
filters -- python list of integers, defining the number of filters in the CONV layers of the main path
|
|
label -- label to use for naming
|
|
s -- Integer, specifying the stride to be used
|
|
|
|
Returns:
|
|
X -- output of the convolutional block, tensor of shape (n_H, n_W, n_C)
|
|
"""
|
|
F1, F2, F3 = filters
|
|
X_shortcut = X
|
|
|
|
# First component of main path
|
|
X = Conv2D(
|
|
F1,
|
|
kernel_size=(1, 1),
|
|
strides=(s,s),
|
|
padding='valid',
|
|
name='conv_%s_2a' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2a' % label)(X)
|
|
X = Activation('relu')(X)
|
|
|
|
# Second component of main path
|
|
X = Conv2D(
|
|
F2,
|
|
kernel_size=(f, f),
|
|
strides=(1, 1),
|
|
padding='same',
|
|
name='conv_%s_2b' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2b' % label)(X)
|
|
X = Activation('relu')(X)
|
|
|
|
# Third component of main path
|
|
X = Conv2D(
|
|
F3,
|
|
kernel_size=(1, 1),
|
|
strides=(1, 1),
|
|
padding='valid',
|
|
name='conv_%s_2c' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X)
|
|
X = BatchNormalization(axis=3, name='bn_%s_2c' % label)(X)
|
|
|
|
# Shortcut path
|
|
X_shortcut = Conv2D(
|
|
F3,
|
|
kernel_size=(1, 1),
|
|
strides=(s, s),
|
|
padding='valid',
|
|
name='conv_%s_1' % label,
|
|
kernel_initializer=glorot_uniform(seed=0)
|
|
)(X_shortcut)
|
|
X_shortcut = BatchNormalization(axis=3, name='bn_%s_1' % label)(X_shortcut)
|
|
|
|
# Add shortcut value to main path, and pass it through a RELU activation
|
|
X = Add()([X_shortcut, X])
|
|
X = Activation('relu')(X)
|
|
|
|
return X
|