def conv_pool(x, kernel_size):
x = Convolution3D(32, (kernel_size, kernel_size, kernel_size),
activation='relu',
padding='same',
data_format='channels_first')(x)
x = MaxPooling3D((3, 3, 3), data_format='channels_first')(x)
return x
def make_model_2(img_channels, img_rows, img_cols,d3, neurons_full_layer):
model = Sequential()
print img_channels, img_rows, img_cols,d3
model.add(Convolution3D(8, 4,4,4, border_mode='same', input_shape=(img_channels, img_rows, img_cols,d3)))
model.add(MaxPooling3D(pool_size=(2, 2,2)))
model.add(Activation('sigmoid'))
#model.add(Convolution2D(64, 3, 3, border_mode='same'))
#model.add(Activation('relu'))
#model.add(Convolution3D(16, 4, 4,4))
#model.add(Activation('sigmoid'))
#model.add(MaxPooling3D(pool_size=(2, 2,2)))
#model.add(Convolution3D(32, 4, 4,4))
#model.add(Activation('relu'))
#model.add(MaxPooling3D(pool_size=(2, 2,2)))
#model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(neurons_full_layer,init='lecun_uniform'))
model.add(Activation('sigmoid'))
#model.add(Dropout(0.20))
model.add(Dense(img_channels,init='lecun_uniform'))
model.add(Activation('linear'))
model.summary()
# let's train the model using SGD + momentum (how original).
opt = Adadelta()
model.compile(loss='mse',optimizer=opt) #, metrics=['accuracy'])
return model
def conv_block(x_input,
num_filters,
pool=True,
activation='relu',
init='orthogonal'):
x1 = Convolution3D(num_filters,
3,
3,
3,
border_mode='same',
W_regularizer=l2(1e-5),
init=init)(x_input)
x1 = BatchNormalization(axis=1, momentum=0.99)(x1)
# x1 = Activation('relu')(x1)
x1 = PReLU(shared_axes=[2, 3, 4], trainable=True)(x1)
#x1 = LeakyReLU(.01)(x1)
#x1 = Convolution3D(num_filters / 2,3,3,3, border_mode='same',W_regularizer=l2(1e-4))(x1)
#x1 = BatchNormalization(axis=1)(x1)
#x1 = LeakyReLU(.1)(x1)
if pool:
x1 = MaxPooling3D()(x1)
x_out = x1
return x_out
def get_model1(self):
nb_filters = [64]
nb_pool = [3]
nb_conv = [7]
# Define model
model = Sequential()
model.add(Convolution3D(
nb_filters[0],
nb_depth=nb_conv[0],
nb_row=nb_conv[0],
nb_col=nb_conv[0],
input_shape=tuple([1] + self.resize),
activation='relu'))
model.add(MaxPooling3D(pool_size=tuple([nb_pool[0]]*3)))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(128, init='normal', activation='relu'))
model.add(Dense(32, init='normal', activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(len(self.labels), init='normal'))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='RMSprop')
return model
def transition_block(input, nb_filter, dropout_rate=None, weight_decay=1E-4):
''' Apply BatchNorm, Relu 1x1, Conv2D, optional dropout and Maxpooling2D
Args:
input: keras tensor
nb_filter: number of filters
dropout_rate: dropout rate
weight_decay: weight decay factor
Returns: keras tensor, after applying batch_norm, relu-conv, dropout, maxpool
'''
concat_axis = 1 if K.image_data_format() == 'channels_first' else -1
x = Convolution3D(nb_filter, (1, 1, 1),
kernel_initializer="he_uniform",
padding="same",
use_bias=False,
kernel_regularizer=l2(weight_decay))(input)
if dropout_rate is not None:
x = Dropout(dropout_rate)(x)
x = AveragePooling3D((2, 2, 2), strides=(2, 2, 2))(x)
x = BatchNormalization(axis=concat_axis,
gamma_regularizer=l2(weight_decay),
beta_regularizer=l2(weight_decay))(x)
return x
def conv_block(x_input,
num_filters,
pool=True,
activation='relu',
init='orthogonal'):
x1 = Convolution3D(num_filters,
3,
3,
3,
border_mode='same',
W_regularizer=l2(1e-5),
init=init)(x_input)
x1 = BatchNormalization(axis=1, momentum=0.99)(x1)
x1 = LeakyReLU(.1)(x1)
#experimental activation kinda like a double sided leaky relu
# x1 = Activation('srelu_fixed')(x1)
# x1 = Convolution3D(num_filters,3,3,3, border_mode='same',W_regularizer=l2(1e-4))(x1)
# x1 = BatchNormalization(axis=1)(x1)
# x1 = LeakyReLU(.1)(x1)
if pool:
x1 = MaxPooling3D()(x1)
x_out = x1
return x_out
def C3D_ucf101_training_model_tf(summary=True, backend='tf'):
""" Return the Keras model of the network
"""
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Flatten
from keras.layers.convolutional import Convolution3D, MaxPooling3D, ZeroPadding3D
from keras.optimizers import SGD
model = Sequential()
if backend == 'tf':
input_shape=(16, 112, 112, 3) # l, h, w, c
else:
input_shape=(3, 16, 112, 112) # c, l, h, w
model.add(Convolution3D(64, 3, 3, 3, activation='relu',
border_mode='same', name='conv1',
input_shape=input_shape))
model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2),
border_mode='valid', name='pool1'))
# 2nd layer group
model.add(Convolution3D(128, 3, 3, 3, activation='relu',
border_mode='same', name='conv2'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool2'))
# 3rd layer group
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3a'))
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool3'))
# 4th layer group
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv4a'))
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv4b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool4'))
# 5th layer group
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5a'))
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5b'))
#model.add(ZeroPadding3D(padding=((0, 0), (0, 1), (0, 1)), name='zeropad5'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool5'))
model.add(Flatten())
# FC layers group
model.add(Dense(4096, activation='relu', name='fc6'))
model.add(Dropout(.5))
model.add(Dense(4096, activation='relu', name='fc7'))
model.add(Dropout(.5))
model.add(Dense(101, activation='linear', name='fc8'))
if summary:
print(model.summary())
return model
def _feed_d_map(self, x, layer_idx):
# shifting down feature maps
x = Lambda(self._shift_depth, name='d_shift_right' + str(layer_idx))(x)
x = Convolution3D(filters=2 * self.nb_filters,
kernel_size=(1, 1, 1),
padding='valid',
name='d_1x1_conv_' + str(layer_idx))(x)
return x
def c3d_model():
classes = 2
""" Return the Keras model of the network
"""
model = Sequential()
input_shape=(16, 100, 100, 3) # l, h, w, c
model.add(Convolution3D(64, 3, 3, 3, activation='relu',
border_mode='same', name='conv1',
input_shape=input_shape))
model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2),
border_mode='valid', name='pool1'))
# 2nd layer group
model.add(Convolution3D(128, 3, 3, 3, activation='relu',
border_mode='same', name='conv2'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool2'))
# 3rd layer group
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3a'))
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool3'))
# 4th layer group
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv4a'))
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv4b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool4'))
# 5th layer group
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5a'))
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool5'))
model.add(Flatten())
# FC layers group
#model.add(Dense(2048, activation='relu', name='fc6'))
#model.add(Dropout(.5))
#model.add(Dense(2048, activation='relu', name='fc7'))
#model.add(Dropout(.5))
#model.add(Dense(2, activation='sigmoid', name='fc8'))
#WS FC Layer with custom parameters
model.add(Dense(512,input_dim=4608,W_regularizer=l2(0.001),activation='relu', name='fc6'))
model.add(Dropout(.6))
model.add(Dense(32,activation='relu',W_regularizer=l2(0.001), name='fc7'))
model.add(Dropout(.6))
model.add(Dense(2, activation='sigmoid',W_regularizer=l2(0.001), name='fc8'))
return model
def c3d_channels(spatial_size,
temporal_size,
classes,
channels,
weights_path=None):
model = Sequential()
model.add(
Convolution3D(filters=64,
kernel_size=(5, 4, 4),
strides=(1, 2, 2),
padding="same",
activation='relu',
name='conv1',
input_shape=(temporal_size, spatial_size, spatial_size,
channels)))
model.add(
MaxPooling3D(pool_size=(2, 2, 2),
strides=(2, 2, 2),
padding="valid",
name='pool1'))
model.add(
Convolution3D(filters=128,
kernel_size=(5, 4, 4),
strides=(1, 2, 2),
padding="same",
activation='relu',
name='conv2'))
model.add(
MaxPooling3D(pool_size=(2, 2, 2),
strides=(2, 2, 2),
padding="valid",
name='pool2'))
model.add(Dense(1024, activation='relu'))
model.add(Dropout(0.25))
model.add(Flatten())
if weights_path:
model.load_weights(weights_path)
model.add(Dense(classes, activation='softmax'))
print(model.summary())
return model
def train_model():
input_img = Input(shape=(128, 128, 128, 1))
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(input_img)
x = MaxPooling3D((2, 2, 2), padding='same')(x)
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(x)
x = MaxPooling3D((2, 2, 2), padding='same')(x)
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(x)
encoded = MaxPooling3D((2, 2, 2), padding='same', name='encoder')(x)
print("shape of encoded: ")
print(K.int_shape(encoded))
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(encoded)
x = UpSampling3D((2, 2, 2))(x)
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(x)
x = UpSampling3D((2, 2, 2))(x)
x = Convolution3D(16, (5, 5, 5), activation='relu', padding='same')(x)
x = UpSampling3D((2, 2, 2))(x)
decoded = Convolution3D(1, (5, 5, 5), activation='sigmoid', padding='same')(x)
print("shape of decoded: ")
print(K.int_shape(decoded))
autoencoder = Model(input_img, decoded)
autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')
autoencoder.fit(train_data, train_data,
epochs=10,
batch_size=14,
validation_data=(val_data, val_data),
callbacks=[TensorBoard(log_dir='/tmp/autoencoder')])
autoencoder.save('autoencoder.h5')
def construct_model(self):
# th input (channels, height, width)
# tf input (height, width, channels)
keras.backend.set_image_dim_ordering('tf')
model = Sequential()
#Encode:
model.add(Convolution3D(16,(2,3,3), activation='relu', input_shape=(config.frames_per_input, 501, 501, 3),border_mode='same'))
model.add(MaxPooling3D(pool_size=(2,2,2)))
model.add(Convolution3D(16,(2,3,3), activation='relu', border_mode='same'))
model.add(Convolution3D(16,(2,3,3), activation='relu', border_mode='same'))
model.add(MaxPooling3D(pool_size=(2,2,2)))
model.add(Convolution3D(8,(1,3,3), activation='relu', border_mode='same'))
#model.add(Convolution3D(8,(1,3,3), activation='relu', border_mode='same'))
#model.add(MaxPooling3D(pool_size=(1,2,2)))
#model.add(Convolution3D(8,(1,3,3), activation='relu', border_mode='same'))
#model.add(Convolution3D(8,(1,3,3), activation='relu', border_mode='same'))
#Decode:
model.add(Convolution3D(16,(2,3,3), activation='relu', border_mode='same'))
model.add(UpSampling3D((1,2,2)))
model.add(Convolution3D(16,(2,3,3), activation='relu', border_mode='same'))
model.add(UpSampling3D((1,2,2)))
model.add(Convolution3D(3,(2,3,3), activation='relu', border_mode='same'))
# model.add(Convolution3D(16,(2,3,3), activation='relu', border_mode='same'))
# model.add(UpSampling3D((1,1,1)))
# model.add(Convolution3D(3,(2,3,3), activation='relu', border_mode='same'))
# model.add(UpSampling3D((1,1,1)))
# model.add(Flatten())
model.compile(loss='mean_squared_error', optimizer='adam')
model.compile(loss=y_std, optimizer='adam')
model.compile(loss='binary_crossentropy', optimizer='adam') # doesn't reset weights
model.summary()
self.model = model
def _feed_v_map(self, x, layer_idx):
# shifting down feature maps
print('v_shift_down' + str(layer_idx))
x = Lambda(self._shift_vert, name='v_shift_down' + str(layer_idx))(x)
x = Convolution3D(filters=2 * self.nb_filters,
kernel_size=(1, 1, 1),
padding='valid',
name='feed_v_1x1x1_conv_' + str(layer_idx))(x)
return x
def seq3DCNN(n_flow=4, seq_len=3, map_height=32, map_width=32):
model=Sequential()
# model.add(ZeroPadding3D(padding=(0, 1, 1), input_shape=(n_flow, seq_len, map_height, map_width)))
# model.add(Convolution3D(64, 2, 3, 3, border_mode='valid'))
model.add(Convolution3D(64, 2, 3, 3, border_mode='same', input_shape=(n_flow, seq_len, map_height, map_width)))
model.add(Activation('relu'))
model.add(Convolution3D(128, 2, 3, 3, border_mode='same'))
model.add(Activation('relu'))
model.add(Convolution3D(64, 2, 3, 3, border_mode='same'))
model.add(Activation('relu'))
model.add(ZeroPadding3D(padding=(0, 1, 1)))
model.add(Convolution3D(n_flow, seq_len, 3, 3, border_mode='valid'))
# model.add(Convolution3D(n_flow, seq_len-2, 3, 3, border_mode='same'))
model.add(Activation('tanh'))
return model
def CNN_convLSTM(activation="relu",
loss="binary_crossentropy",
optimizer="Adadelta",
layer=0,
height=0,
width=0,
days=0,
timesteps=0):
"""
INPUT -> [CONV -> RELU] -> OUT
"""
model = Sequential()
# model.add(ZeroPadding3D((1, 1, 1), data_format="channels_last",
# input_shape=(timesteps, height, width, layer)))
# model.add(BatchNormalization())
model.add(
ConvLSTM2D(filters=16,
kernel_size=(3, 3),
padding='same',
activation=activation,
data_format="channels_last",
input_shape=(None, height, width, layer),
return_sequences=True))
model.add(BatchNormalization())
model.add(
ConvLSTM2D(filters=16,
kernel_size=(3, 3),
padding='same',
activation=activation,
data_format="channels_last",
return_sequences=True))
model.add(BatchNormalization())
model.add(
ConvLSTM2D(filters=32,
kernel_size=(3, 3),
padding='same',
activation=activation,
data_format="channels_last",
return_sequences=True))
model.add(BatchNormalization())
model.add(
ConvLSTM2D(filters=64,
kernel_size=(3, 3),
padding='same',
activation=activation,
data_format="channels_last",
return_sequences=True))
model.add(BatchNormalization())
model.add(
Convolution3D(filters=layer,
kernel_size=(3, 3, layer),
padding="same",
data_format="channels_last"))
model.compile(loss=loss, optimizer=optimizer)
return model
def conv2_block(input, k=1, dropout=0.0):
init = input
# Check if input number of filters is same as 32 * k, else create convolution2d for this input
if init._keras_shape[1] != 32 * k:
init = Convolution3D(32 * k,
1,
1,
1,
activation='linear',
border_mode='same',
init=weight_init,
W_regularizer=l2(weight_decay),
bias=use_bias)(init)
x = Convolution3D(32 * k,
3,
3,
3,
border_mode='same',
init=weight_init,
W_regularizer=l2(weight_decay),
bias=use_bias)(input)
x = BatchNormalization(axis=1)(x)
# x = Activation('relu')(x)
x = LeakyReLU(0.3)(x)
if dropout > 0.0:
x = Dropout(dropout)(x)
x = Convolution3D(32 * k,
3,
3,
3,
border_mode='same',
init=weight_init,
W_regularizer=l2(weight_decay),
bias=use_bias)(x)
x = BatchNormalization(axis=1)(x)
# x = Activation('relu')(x)
x = LeakyReLU(0.3)(x)
m = merge([init, x], mode='sum')
return m
def f(input):
activation = _bn_relu(input)
return Convolution3D(nb_filter=nb_filter,
kernel_dim1=kernel_dim1,
kernel_dim2=kernel_dim2,
kernel_dim3=kernel_dim3,
subsample=subsample,
init=init,
border_mode=border_mode,
W_regularizer=W_regularizer)(activation)
def get_model(summary=False, backend='tf'):
""" Return the Keras model of the network
"""
model = Sequential()
if backend == 'tf':
input_shape=(16, 112, 112, 3) # l, h, w, c
else:
input_shape=(3, 16, 112, 112) # c, l, h, w
model.add(Convolution3D(64, 3, 3, 3, activation='relu',
border_mode='same', name='conv1',
input_shape=input_shape))
model.add(MaxPooling3D(pool_size=(1, 2, 2), strides=(1, 2, 2),
border_mode='valid', name='pool1'))
# 2nd layer group
model.add(Convolution3D(128, 3, 3, 3, activation='relu',
border_mode='same', name='conv2'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool2'))
# 3rd layer group
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3a'))
model.add(Convolution3D(256, 3, 3, 3, activation='relu',
border_mode='same', name='conv3b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool3'))
# 4th layer group
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv4a'))
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv4b'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool4'))
# 5th layer group
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5a'))
model.add(Convolution3D(512, 3, 3, 3, activation='relu',
border_mode='same', name='conv5b'))
model.add(ZeroPadding3D(padding=(0, 1, 1), name='zeropad5'))
model.add(MaxPooling3D(pool_size=(2, 2, 2), strides=(2, 2, 2),
border_mode='valid', name='pool5'))
model.add(Flatten())
# FC layers group
model.add(Dense(4096, activation='relu', name='fc6'))
model.add(Dropout(.5))
model.add(Dense(4096, activation='relu', name='fc7'))
model.add(Dropout(.5))
model.add(Dense(487, activation='softmax', name='fc8'))
if summary:
print(model.summary())
return model
