def get_drop_block_model():
model = keras.models.Sequential()
model.add(
DropBlock2D(input_shape=(28, 28, 1),
block_size=7,
keep_prob=0.8,
name='Input-Dropout'))
model.add(
keras.layers.Conv2D(filters=64,
kernel_size=3,
activation='relu',
padding='same',
name='Conv-1'))
model.add(keras.layers.MaxPool2D(pool_size=2, name='Pool-1'))
model.add(DropBlock2D(block_size=5, keep_prob=0.8, name='Dropout-1'))
model.add(
keras.layers.Conv2D(filters=32,
kernel_size=3,
activation='relu',
padding='same',
name='Conv-2'))
model.add(keras.layers.MaxPool2D(pool_size=2, name='Pool-2'))
model.add(DropBlock2D(block_size=3, keep_prob=0.8, name='Dropout-2'))
model.add(keras.layers.Flatten(name='Flatten'))
model.add(keras.layers.Dense(units=256, activation='relu', name='Dense'))
model.add(keras.layers.Dropout(rate=0.2, name='Dense-Dropout'))
model.add(
keras.layers.Dense(units=10, activation='softmax', name='Softmax'))
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'],
)
return model
def test_mask_shape(self):
input_layer = keras.layers.Input(shape=(10, 10, 3))
drop_block_layer = DropBlock2D(block_size=3,
keep_prob=0.7)(input_layer,
training=True)
model = keras.models.Model(inputs=input_layer,
outputs=drop_block_layer)
model.compile(optimizer='adam', loss='mse', metrics={})
model_path = os.path.join(tempfile.gettempdir(),
'keras_drop_block_%f.h5' % random.random())
model.save(model_path)
model = keras.models.load_model(
model_path,
custom_objects={'DropBlock2D': DropBlock2D},
)
model.summary()
inputs = np.ones((1, 10, 10, 3))
outputs = model.predict(inputs)
for i in range(3):
print((outputs[0, :, :, i] > 0.0).astype(dtype='int32'))
inputs = np.ones((1000, 10, 10, 3))
outputs = model.predict(inputs)
keep_prob = 1.0 * np.sum(outputs > 0.0) / np.prod(np.shape(outputs))
print(keep_prob)
self.assertTrue(0.65 < keep_prob < 0.8, keep_prob)
input_layer = keras.layers.Input(shape=(3, 10, 10))
drop_block_layer = DropBlock2D(block_size=3,
keep_prob=0.7,
data_format='channels_first')(
input_layer, training=True)
model = keras.models.Model(inputs=input_layer,
outputs=drop_block_layer)
model.compile(optimizer='adam', loss='mse', metrics={})
model_path = os.path.join(tempfile.gettempdir(),
'keras_drop_block_%f.h5' % random.random())
model.save(model_path)
model = keras.models.load_model(
model_path,
custom_objects={'DropBlock2D': DropBlock2D},
)
model.summary()
inputs = np.ones((1, 3, 10, 10))
outputs = model.predict(inputs)
for i in range(3):
print((outputs[0, i, :, :] > 0.0).astype(dtype='int32'))
inputs = np.ones((1000, 3, 10, 10))
outputs = model.predict(inputs)
keep_prob = 1.0 * np.sum(outputs > 0.0) / np.prod(np.shape(outputs))
print(keep_prob)
self.assertTrue(0.65 < keep_prob < 0.8, keep_prob)
def test_sync_channels(self):
input_layer = keras.layers.Input(shape=(10, 10, 3))
drop_block_layer = keras.layers.Lambda(
lambda x: DropBlock2D(
block_size=3, keep_prob=0.7, sync_channels=True)
(x, training=True), )(input_layer)
model = keras.models.Model(inputs=input_layer,
outputs=drop_block_layer)
model.compile(optimizer='adam', loss='mse', metrics={})
model_path = os.path.join(tempfile.gettempdir(),
'keras_drop_block_%f.h5' % random.random())
model.save(model_path)
model = keras.models.load_model(
model_path,
custom_objects={'DropBlock2D': DropBlock2D},
)
model.summary()
inputs = np.ones((1, 10, 10, 3))
outputs = model.predict(inputs)
for i in range(1, 3):
self.assertTrue(
np.allclose(outputs[0, :, :, 0], outputs[0, :, :, i]))
inputs = np.ones((1000, 10, 10, 3))
outputs = model.predict(inputs)
keep_prob = 1.0 * np.sum(outputs > 0.0) / np.prod(np.shape(outputs))
self.assertTrue(0.6 < keep_prob < 0.8, keep_prob)
def decoder(self, features=[8], name="decoder") -> KM.Model:
"""Creates a decoder model object
Args:
features (list, optional): list of features in successive hidden layers. Defaults to [8].
name (str, optional): name for the model object. Defaults to "decoder".
Returns:
KM.Model: Decoder model
"""
input_tensor = KL.Input(shape=(2, 2, features[0]))
decoded = input_tensor
for i, feature_num in enumerate(features[1:], start=1):
decoded = deconv_block(decoded, feature_num,
name + f"_deconv_{len(features)-i}")
# Final reconstruction back to the original image size
decoded = KL.Conv2DTranspose(
3,
(4, 4),
strides=(2, 2),
padding="same",
kernel_initializer="he_normal",
use_bias=True,
activation="tanh",
name=name + f"_out",
)(decoded)
decoded = DropBlock2D(block_size=5, keep_prob=0.8)(decoded)
return KM.Model(inputs=input_tensor, outputs=decoded, name=name)
def test_training(self):
input_layer = keras.layers.Input(shape=(10, 10, 3))
drop_block_layer = DropBlock2D(block_size=3,
keep_prob=0.7)(input_layer)
model = keras.models.Model(inputs=input_layer,
outputs=drop_block_layer)
model.compile(optimizer='adam', loss='mse', metrics={})
model_path = os.path.join(tempfile.gettempdir(),
'keras_drop_block_%f.h5' % random.random())
model.save(model_path)
model = keras.models.load_model(
model_path,
custom_objects={'DropBlock2D': DropBlock2D},
)
model.summary()
inputs = np.ones((1, 10, 10, 3))
outputs = model.predict(inputs)
self.assertTrue(np.allclose(inputs, outputs))
input_layer = keras.layers.Input(shape=(3, 10, 10))
drop_block_layer = DropBlock2D(
block_size=3, keep_prob=0.7,
data_format='channels_first')(input_layer)
model = keras.models.Model(inputs=input_layer,
outputs=drop_block_layer)
model.compile(optimizer='adam', loss='mse', metrics={})
model_path = os.path.join(tempfile.gettempdir(),
'keras_drop_block_%f.h5' % random.random())
model.save(model_path)
model = keras.models.load_model(
model_path,
custom_objects={'DropBlock2D': DropBlock2D},
)
model.summary()
inputs = np.ones((1, 3, 10, 10))
outputs = model.predict(inputs)
self.assertTrue(np.allclose(inputs, outputs))
y_train, y_test = np.expand_dims(y_train, axis=-1), np.expand_dims(y_test, axis=-1) train_num = round(x_train.shape[0]) x_train, x_valid = x_train[:train_num, ...], x_train[train_num:, ...] y_train, y_valid = y_train[:train_num, ...], y_train[train_num:, ...] # Design the model model = Sequential() model.add(ZeroPadding2D(input_shape=(28, 28, 1), name='Input')) model.add(Conv2D(filters=32, strides=(1, 1), kernel_size=3, activation='relu', padding='same', name='Conv-1')) model.add(BatchNormalization()) model.add(Conv2D(filters=32, strides=(1, 1), kernel_size=3, activation='relu', padding='same', name='Conv-2')) model.add(BatchNormalization()) model.add(MaxPool2D(pool_size=2, name='Pool-1')) model.add(DropBlock2D(block_size=7, keep_prob=0.8, name='Dropout-1')) model.add(Conv2D(filters=64, strides=(1, 1), kernel_size=3, activation='relu', padding='same', name='Conv-3')) model.add(BatchNormalization()) model.add(MaxPool2D(pool_size=2, name='Pool-2')) model.add(DropBlock2D(block_size=7, keep_prob=0.8, name='Dropout-2')) model.add(Conv2D(filters=64, strides=(1, 1), kernel_size=3, activation='relu', padding='same', name='Conv-4')) model.add(BatchNormalization()) model.add(MaxPool2D(pool_size=2, name='Pool-3')) model.add(DropBlock2D(block_size=7, keep_prob=0.8, name='Dropout-3')) model.add(Conv2D(filters=128, strides=(1, 1), kernel_size=3, activation='relu', padding='same', name='Conv-5')) model.add(BatchNormalization()) model.add(MaxPool2D(pool_size=2, name='Pool-4')) model.add(DropBlock2D(block_size=7, keep_prob=0.8, name='Dropout-4'))
x_train, x_test, y_train, y_test = train_test_split(data,
labels,
test_size=0.3)
base_model = VGG16(weights='imagenet',
include_top=False,
input_shape=(32, 32, 3),
pooling='avg')
b2 = Model(inputs=base_model.input,
outputs=base_model.get_layer('block3_pool').output)
fc1 = b2.layers[-3]
fc2 = b2.layers[-2]
predictions = b2.layers[-1]
dropout1 = DropBlock2D(block_size=3, keep_prob=0.8)
dropout2 = DropBlock2D(block_size=3, keep_prob=0.8)
x = dropout1(fc1.output)
x = fc2(x)
x = dropout2(x)
x = predictions(x)
x = Flatten()(x)
vgg_conv = Model(inputs=base_model.input, outputs=x)
for layer in vgg_conv.layers:
layer.trainable = False
vgg_conv.layers[-3].trainable = True
vgg_conv.layers[-4].trainable = True
vgg_conv.layers[-5].trainable = True
vgg_conv.summary()
def identity_block(X, f, filters, stage, block):
conv_name_base = 'res' + str(stage) + block + '_branch'
bn_name_base = 'bn' + str(stage) + block + '_branch'
# Retrieve Filters
F1, F2, F3 = filters
# Save the input value. You'll need this later to add back to the main path.
X_shortcut = X
if stage == 4:
dilation = 2
elif stage == 5:
if block == 'b' or block == 'c':
dilation = 4
elif block == 'd':
dilation = 2
else:
dilation = 1
else:
dilation = 1
# First component of main path
X = AtrousConvolution2D(filters=F1,
kernel_size=(1, 1),
strides=(1, 1),
padding='valid',
name=conv_name_base + '2a',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2a')(X)
X = Activation('relu')(X)
# Second component of main path (≈3 lines)
X = AtrousConvolution2D(filters=F2,
kernel_size=(f, f),
atrous_rate=(dilation, dilation),
padding='same',
name=conv_name_base + '2b',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2b')(X)
X = Activation('relu')(X)
# Third component of main path (≈2 lines)
X = AtrousConvolution2D(filters=F3,
kernel_size=(1, 1),
strides=(1, 1),
padding='valid',
name=conv_name_base + '2c',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2c')(X)
if stage == 5:
if block == 'e':
X = Add()([X, X_shortcut])
X = Activation('relu')(X)
else:
X = Activation('relu')(X)
return X
def convolutional_block(X, f, filters, stage, block, s=2):
conv_name_base = 'res' + str(stage) + block + '_branch'
bn_name_base = 'bn' + str(stage) + block + '_branch'
# Retrieve Filters
F1, F2, F3 = filters
# Save the input value
X_shortcut = X
if stage == 4 or stage == 5:
dilation = 2
else:
dilation = 1
##### MAIN PATH #####
# First component of main path
X = AtrousConvolution2D(F1, (1, 1),
strides=(s, s),
name=conv_name_base + '2a',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2a')(X)
X = Activation('relu')(X)
# Second component of main path (≈3 lines)
X = AtrousConvolution2D(filters=F2,
kernel_size=(f, f),
atrous_rate=(dilation, dilation),
padding='same',
name=conv_name_base + '2b',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2b')(X)
X = Activation('relu')(X)
# Third component of main path (≈2 lines)
X = AtrousConvolution2D(filters=F3,
kernel_size=(1, 1),
strides=(1, 1),
padding='valid',
name=conv_name_base + '2c',
kernel_initializer=glorot_uniform(seed=0))(X)
if stage == 4 or stage == 5:
X = DropBlock2D(block_size=7, keep_prob=0.9)(X)
X = BatchNormalization(axis=3, name=bn_name_base + '2c')(X)
##### SHORTCUT PATH #### (≈2 lines)
X_shortcut = AtrousConvolution2D(
filters=F3,
kernel_size=(1, 1),
strides=(s, s),
padding='valid',
name=conv_name_base + '1',
kernel_initializer=glorot_uniform(seed=0))(X_shortcut)
if stage == 4 or stage == 5:
X_shortcut = DropBlock2D(block_size=7, keep_prob=0.9)(X_shortcut)
X_shortcut = BatchNormalization(axis=3,
name=bn_name_base + '1')(X_shortcut)
# Final step: Add shortcut value to main path, and pass it through a RELU activation (≈2 lines)
X = Add()([X, X_shortcut])
X = Activation('relu')(X)
return X