kernel_size=(3, 3),
activation='relu',
name='block4_conv2')(h)
h = ZeroPadding2D(padding=(1, 1))(h)
h = Conv2D(filters=512,
kernel_size=(3, 3),
activation='relu',
name='block4_conv3')(h)
h = ZeroPadding2D(padding=(1, 1))(h)
h = MaxPool2D(pool_size=(3, 3), strides=(1, 1))(h)
# block5 (start to use Atrous convolution)
h = ZeroPadding2D(padding=(2, 2))(h) # padding <-> atrous_rate
h = AtrousConv2D(filters=512,
atrous_rate=(2, 2),
kernel_size=(3, 3),
activation='relu',
name='block5_conv1')(h)
h = ZeroPadding2D(padding=(2, 2))(h)
h = AtrousConv2D(filters=512,
atrous_rate=(2, 2),
kernel_size=(3, 3),
activation='relu',
name='block5_conv2')(h)
h = ZeroPadding2D(padding=(2, 2))(h)
h = AtrousConv2D(filters=512,
atrous_rate=(2, 2),
kernel_size=(3, 3),
activation='relu',
name='block5_conv3')(h)
h = ZeroPadding2D(padding=(1, 1))(h)
if args.model == "conv":
model = Sequential()
model.add(Conv2D(8, kernel_size=(3, 3),
activation="relu",
input_shape=input_shape))
model.add(Conv2D(16, (3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(32, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
elif args.model == "dilated_conv":
model = Sequential()
model.add(AtrousConv2D(8, kernel_size=(3, 3), atrous_rate=(2, 2),
activation="relu",
input_shape=input_shape)) # shape is 5x5
model.add(AtrousConv2D(16, kernel_size=(3, 3), atrous_rate=(3, 3),
activation="relu")) # shape is 7x7
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(32, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
elif args.model == "fc":
model = Sequential()
model.add(Dense(512, activation="hard_sigmoid", input_shape=(784,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation="relu"))
model.add(Dropout(0.2))
def _setup_model(model_type):
from keras import backend as K
from keras.layers import Dense, Dropout, Flatten, Conv2D, AtrousConv2D, MaxPooling2D, Input, add, GlobalAveragePooling2D, Activation
from keras.models import Sequential, Model
input_shape = get_input_shape(model_type)
if model_type == "conv":
model = Sequential()
model.add(
Conv2D(8,
kernel_size=(3, 3),
activation="relu",
input_shape=input_shape))
model.add(Conv2D(16, (3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(32, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
elif model_type == "dilated_conv":
model = Sequential()
model.add(
AtrousConv2D(8,
kernel_size=(3, 3),
atrous_rate=(2, 2),
activation="relu",
input_shape=input_shape)) # shape is 5x5
model.add(
AtrousConv2D(16,
kernel_size=(3, 3),
atrous_rate=(3, 3),
activation="relu")) # shape is 7x7
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(32, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation="softmax"))
elif model_type == "fc":
model = Sequential()
model.add(Dense(512, activation="hard_sigmoid", input_shape=(784, )))
model.add(Dropout(0.2))
model.add(Dense(512, activation="relu"))
model.add(Dropout(0.2))
model.add(Dense(10, activation="softmax"))
elif model_type == "residual":
nn_input = Input(shape=(28, 28, 1))
hidden = Conv2D(8, kernel_size=(3, 3), activation="relu")(nn_input)
hidden = MaxPooling2D(pool_size=(2, 2))(hidden)
hidden_1 = Conv2D(16,
kernel_size=(1, 1),
activation="relu",
padding="same")(hidden)
hidden_2 = Conv2D(16,
kernel_size=(3, 3),
activation="relu",
padding="same")(hidden)
hidden = add([hidden_1, hidden_2])
hidden_1 = hidden
hidden_2 = Conv2D(16,
kernel_size=(3, 3),
activation="relu",
padding="same")(hidden)
hidden = add([hidden_1, hidden_2])
hidden = GlobalAveragePooling2D()(hidden)
nn_output = Dense(num_classes, activation="softmax")(hidden)
model = Model(inputs=[nn_input], outputs=[nn_output])
elif model_type == "complex":
# graph which has graph and sequential
# this is for testing converting complex model
nn_input = Input(shape=(28, 28, 1))
hidden_1 = Conv2D(8, kernel_size=(3, 3), activation="relu")(nn_input)
submodel_input = Input(shape=(26, 26, 8))
submodel_conv = Conv2D(8, kernel_size=(3, 3), activation="relu")
submodel_1 = submodel_conv(submodel_input)
submodel_2 = submodel_conv(submodel_1) # use same layer multiple times
submodel_3 = Conv2D(16, kernel_size=(3, 3),
activation="relu")(submodel_1)
submodel = Model(inputs=[submodel_input],
outputs=[submodel_3, submodel_2])
subseq = Sequential()
subseq.add(
Conv2D(16,
kernel_size=(3, 3),
activation="relu",
input_shape=(22, 22, 16)))
subseq.add(Flatten())
subseq.add(Dense(10))
hidden_2, hidden_3 = submodel(hidden_1)
hidden_4 = subseq(hidden_2)
hidden_5 = Flatten()(hidden_3)
hidden_6 = Dense(10)(hidden_5)
hidden_sum = add([hidden_4, hidden_6])
nn_output = Activation(activation="softmax")(hidden_sum)
model = Model(inputs=[nn_input], outputs=[nn_output])
else:
raise NotImplementedError("Unknown model type")
print(f"input shape: {input_shape}, data_format: {K.image_data_format()}")
return model
def VGG19Dense(weights='imagenet', input_tensor=None):
'''Instantiate the VGG19 architecture,
optionally loading weights pre-trained
on ImageNet. Note that when using TensorFlow,
for best performance you should set
`image_dim_ordering="tf"` in your Keras config
at ~/.keras/keras.json.
The model and the weights are compatible with both
TensorFlow and Theano. The dimension ordering
convention used by the model is the one
specified in your Keras config file.
# Arguments
weights: one of `None` (random initialization)
or "imagenet" (pre-training on ImageNet).
input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
to use as image input for the model.
# Returns
A Keras model instance.
'''
if weights not in {'imagenet', None}:
raise ValueError('The `weights` argument should be either '
'`None` (random initialization) or `imagenet` '
'(pre-training on ImageNet).')
# Determine proper input shape
if K.image_dim_ordering() == 'th':
input_shape = (3, None, None)
else:
input_shape = (None, None, 3)
if input_tensor is None:
img_input = Input(shape=input_shape)
else:
if not K.is_keras_tensor(input_tensor):
img_input = Input(tensor=input_tensor)
else:
img_input = input_tensor
# Block 1
x = AtrousConv2D(64, 3, 3, atrous_rate=(1,1), activation='relu', border_mode='same', name='block1_conv1')(img_input)
x = AtrousConv2D(64, 3, 3, atrous_rate=(1,1), activation='relu', border_mode='same', name='block1_conv2')(x)
#x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
# Block 2
x = AtrousConv2D(128, 3, 3, atrous_rate=(2,2), activation='relu', border_mode='same', name='block2_conv1')(x)
x = AtrousConv2D(128, 3, 3, atrous_rate=(2,2), activation='relu', border_mode='same', name='block2_conv2')(x)
#x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)
# Block 3
x = AtrousConv2D(256, 3, 3, atrous_rate=(4,4), activation='relu', border_mode='same', name='block3_conv1')(x)
x = AtrousConv2D(256, 3, 3, atrous_rate=(4,4), activation='relu', border_mode='same', name='block3_conv2')(x)
x = AtrousConv2D(256, 3, 3, atrous_rate=(4,4), activation='relu', border_mode='same', name='block3_conv3')(x)
x = AtrousConv2D(256, 3, 3, atrous_rate=(4,4), activation='relu', border_mode='same', name='block3_conv4')(x)
#x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool')(x)
# Block 4
x = AtrousConv2D(512, 3, 3, atrous_rate=(8,8), activation='relu', border_mode='same', name='block4_conv1')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(8,8), activation='relu', border_mode='same', name='block4_conv2')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(8,8), activation='relu', border_mode='same', name='block4_conv3')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(8,8), activation='relu', border_mode='same', name='block4_conv4')(x)
#x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool')(x)
# Block 5
x = AtrousConv2D(512, 3, 3, atrous_rate=(16,16), activation='relu', border_mode='same', name='block5_conv1')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(16,16), activation='relu', border_mode='same', name='block5_conv2')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(16,16), activation='relu', border_mode='same', name='block5_conv3')(x)
x = AtrousConv2D(512, 3, 3, atrous_rate=(16,16), activation='relu', border_mode='same', name='block5_conv4')(x)
#x = MaxPooling2D((2, 2), strides=(2, 2), name='block5_pool')(x)
# Create model
model = Model(img_input, x)
# load weights
if weights == 'imagenet':
print('K.image_dim_ordering:', K.image_dim_ordering())
if K.image_dim_ordering() == 'th':
weights_path = get_file('vgg19_weights_th_dim_ordering_th_kernels_notop.h5',
TH_WEIGHTS_PATH_NO_TOP,
cache_subdir='models')
model.load_weights(weights_path)
if K.backend() == 'tensorflow':
warnings.warn('You are using the TensorFlow backend, yet you '
'are using the Theano '
'image dimension ordering convention '
'(`image_dim_ordering="th"`). '
'For best performance, set '
'`image_dim_ordering="tf"` in '
'your Keras config '
'at ~/.keras/keras.json.')
convert_all_kernels_in_model(model)
else:
weights_path = get_file('vgg19_weights_tf_dim_ordering_tf_kernels_notop.h5',
TF_WEIGHTS_PATH_NO_TOP,
cache_subdir='models')
model.load_weights(weights_path)
if K.backend() == 'theano':
convert_all_kernels_in_model(model)
return model