training_data = training_data_generator.flow_from_directory(
"/content/drive/My Drive/Colabbar Tags Data/Images",
batch_size=32,
target_size=(300, 300),
subset="training",
class_mode='categorical')
validation_data = training_data_generator.flow_from_directory(
"/content/drive/My Drive/Colabbar Tags Data/Images",
batch_size=32,
target_size=(300, 300),
subset="validation",
class_mode='categorical')
#Model definition
model = Sequential()
model.add(Conv2D(filters=16, kernel_size=(5, 5), activation="relu", input_shape=(300, 300, 3)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Conv2D(filters=32, kernel_size=(5, 5), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Conv2D(filters=64, kernel_size=(5, 5), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Conv2D(filters=64, kernel_size=(5, 5), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64, activation='relu'))
def puigcerver_octconv(input_size, d_model):
"""
Octave CNN by khinggan, architecture is same as puigcerver
"""
alpha = 0.25
input_data = Input(name="input", shape=input_size)
high = input_data
low = tf.keras.layers.AveragePooling2D(2)(input_data)
high, low = OctConv2D(filters=16, alpha=alpha)([high, low])
high = BatchNormalization()(high)
low = BatchNormalization()(low)
high = LeakyReLU(alpha=0.01)(high)
low = LeakyReLU(alpha=0.01)(low)
high = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding="valid")(high)
low = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(low)
high, low = OctConv2D(filters=32, alpha=alpha)([high, low])
high = BatchNormalization()(high)
low = BatchNormalization()(low)
high = LeakyReLU(alpha=0.01)(high)
low = LeakyReLU(alpha=0.01)(low)
high = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding="valid")(high)
low = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(low)
high = Dropout(rate=0.2)(high)
low = Dropout(rate=0.2)(low)
high = Conv2D(filters=48,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(high)
low = Conv2D(filters=48,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(low)
high = BatchNormalization()(high)
low = BatchNormalization()(low)
high = LeakyReLU(alpha=0.01)(high)
low = LeakyReLU(alpha=0.01)(low)
high = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding="valid")(high)
low = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(low)
high = Dropout(rate=0.2)(high)
low = Dropout(rate=0.2)(low)
high = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(high)
low = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(low)
high = BatchNormalization()(high)
low = BatchNormalization()(low)
high = LeakyReLU(alpha=0.01)(high)
low = LeakyReLU(alpha=0.01)(low)
high = Dropout(rate=0.2)(high)
low = Dropout(rate=0.2)(low)
high = Conv2D(filters=80,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(high)
low = Conv2D(filters=80,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(low)
high = BatchNormalization()(high)
low = BatchNormalization()(low)
high = LeakyReLU(alpha=0.01)(high)
low = LeakyReLU(alpha=0.01)(low)
x = _create_octconv_last_block([high, low], 80, alpha)
shape = x.get_shape()
blstm = Reshape((shape[1], shape[2] * shape[3]))(x)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Dropout(rate=0.5)(blstm)
output_data = Dense(units=d_model, activation="softmax")(blstm)
return (input_data, output_data)
from keras.layers.normalization import BatchNormalization
#Specifying input image shape
image_shape = (227, 227, 3)
#instantiating an empty model
np.random.seed(1000)
#Ceating the AlexNet DNN
model = Sequential([
Conv2D(filters=96,
kernel_size=(11, 11),
strides=(4, 4),
input_shape=(image_shape),
activation="relu",
padding='valid'),
MaxPooling2D(pool_size=(3, 3), padding='valid', strides=(2, 2)),
Conv2D(256, (5, 5), strides=(1, 1), padding='same', activation="relu"),
MaxPooling2D(pool_size=(3, 3), strides=(2, 2)),
Conv2D(384, (3, 3), strides=(1, 1), padding='same', activation="relu"),
Conv2D(384, (3, 3), strides=(1, 1), padding='same', activation="relu"),
Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation="relu"),
MaxPooling2D(pool_size=(3, 3), strides=(2, 2)),
Flatten(),
Dense(units=4096, activation="relu"),
Dropout(rate=0.5),
Dense(units=4096, activation="relu"),
Dropout(rate=0.5),
Dense(units=1000, activation="softmax"),
import numpy as np
import cv2
import os
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten
from tensorflow.keras.layers import Conv2D
from tensorflow.keras.layers import MaxPooling2D
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
tf_model = Sequential()
tf_model.add(
Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(48, 48, 1)))
tf_model.add(Conv2D(64, kernel_size=(3, 3), activation='relu'))
tf_model.add(MaxPooling2D(pool_size=(2, 2)))
tf_model.add(Dropout(0.25))
tf_model.add(Conv2D(128, kernel_size=(3, 3), activation='relu'))
tf_model.add(MaxPooling2D(pool_size=(2, 2)))
tf_model.add(Conv2D(128, kernel_size=(3, 3), activation='relu'))
tf_model.add(MaxPooling2D(pool_size=(2, 2)))
tf_model.add(Dropout(0.25))
tf_model.add(Flatten())
tf_model.add(Dense(1024, activation='relu'))
tf_model.add(Dropout(0.5))
tf_model.add(Dense(7, activation='softmax'))
tf_model.load_weights('ai_data/model.h5')
facecasc = cv2.CascadeClassifier('ai_data/haarcascade_frontalface_default.xml')
def puigcerver(input_size, d_model):
"""
Convolucional Recurrent Neural Network by Puigcerver et al.
Reference:
Joan Puigcerver.
Are multidimensional recurrent layers really necessary for handwritten text recognition?
In: Document Analysis and Recognition (ICDAR), 2017 14th
IAPR International Conference on, vol. 1, pp. 67–72. IEEE (2017)
Carlos Mocholí Calvo and Enrique Vidal Ruiz.
Development and experimentation of a deep learning system for convolutional and recurrent neural networks
Escola Tècnica Superior d’Enginyeria Informàtica, Universitat Politècnica de València, 2018
"""
input_data = Input(name="input", shape=input_size)
cnn = Conv2D(filters=16,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(input_data)
cnn = BatchNormalization()(cnn)
cnn = LeakyReLU(alpha=0.01)(cnn)
cnn = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(cnn)
cnn = Conv2D(filters=32,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = BatchNormalization()(cnn)
cnn = LeakyReLU(alpha=0.01)(cnn)
cnn = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=48,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = BatchNormalization()(cnn)
cnn = LeakyReLU(alpha=0.01)(cnn)
cnn = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding="valid")(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = BatchNormalization()(cnn)
cnn = LeakyReLU(alpha=0.01)(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=80,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = BatchNormalization()(cnn)
cnn = LeakyReLU(alpha=0.01)(cnn)
shape = cnn.get_shape()
blstm = Reshape((shape[1], shape[2] * shape[3]))(cnn)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Bidirectional(LSTM(units=256, return_sequences=True,
dropout=0.5))(blstm)
blstm = Dropout(rate=0.5)(blstm)
output_data = Dense(units=d_model, activation="softmax")(blstm)
return (input_data, output_data)
# Creating the Test set
valid_set = test_datagen.flow_from_directory('images/validation',
target_size = (48,48),
batch_size = 32,
color_mode='grayscale',
class_mode = 'categorical')
# number of possible label values
n = 7
#print(dir(trainig_set))
# Initialising the CNN
model = Sequential()
# 1 - Convolution
model.add(Conv2D(32,(3,3), padding='same', input_shape=(48, 48,1)))
model.add(Activation('relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
# 2nd Convolution layer
model.add(Conv2D(64,(5,5), padding='same'))
model.add(Activation('relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
# 3rd Convolution layer
model.add(Conv2D(128,(3,3), padding='same'))
model.add(Activation('relu'))
def build(self, input_shapes):
self.conv = Conv2D(filters=1,
kernel_size=self.size,
strides=1,
padding='same')
def DeepSTN(
H=32,
W=32,
channel=2, #H-map_height W-map_width channel-map_channel
c=3,
p=1,
t=1, #c-closeness p-period t-trend
pre_F=64,
conv_F=64,
R_N=2, #pre_F-prepare_conv_featrue conv_F-resnet_conv_featrue R_N-resnet_number
is_plus=True, #use ResPlus or mornal convolution
is_plus_efficient=False, #use the efficient version of ResPlus
plus=8,
rate=2, #rate-pooling_rate
is_pt=True, #use PoI and Time or not
P_N=9,
T_F=31,
PT_F=9,
T=24, #P_N-poi_number T_F-time_feature PT_F-poi_time_feature T-T_times/day
drop=0,
is_summary=True, #show detail
lr=0.0002,
kernel1=1, #kernel1 decides whether early-fusion uses conv_unit0 or conv_unit1, 1 recommended
isPT_F=1
): #isPT_F decides whether PT_model uses one more Conv after multiplying PoI and Time, 1 recommended
all_channel = channel * (c + p + t)
cut0 = int(0)
cut1 = int(cut0 + channel * c)
cut2 = int(cut1 + channel * p)
cut3 = int(cut2 + channel * t)
cpt_input = Input(shape=(H, W, all_channel))
c_input = Lambda(cpt_slice, arguments={'h1': cut0, 'h2': cut1})(cpt_input)
p_input = Lambda(cpt_slice, arguments={'h1': cut1, 'h2': cut2})(cpt_input)
t_input = Lambda(cpt_slice, arguments={'h1': cut2, 'h2': cut3})(cpt_input)
c_out1 = Conv2D(filters=pre_F, kernel_size=(1, 1), padding="same")(c_input)
p_out1 = Conv2D(filters=pre_F, kernel_size=(1, 1), padding="same")(p_input)
t_out1 = Conv2D(filters=pre_F, kernel_size=(1, 1), padding="same")(t_input)
# print(t_out1.shape)
if is_pt:
poi_in = Input(shape=(P_N, H, W))
# T_times/day + 7days/week
time_in = Input(shape=(T + 7, H, W))
PT_model = PT_trans('PT_trans', P_N, PT_F, T, T_F, H, W, isPT_F)
poi_time = PT_model([poi_in, time_in])
cpt_con1 = Concatenate(axis=-1)([c_out1, p_out1, t_out1, poi_time])
# print("="*10)
# print(c_out1.shape,p_out1.shape,t_out1.shape,poi_time.shape)
# print(cpt_con1.shape)
if kernel1:
cpt = conv_unit1(pre_F * 3 + PT_F * isPT_F + P_N * (not isPT_F),
conv_F, drop, H, W)(cpt_con1)
else:
cpt = conv_unit0(pre_F * 3 + PT_F * isPT_F + P_N * (not isPT_F),
conv_F, drop, H, W)(cpt_con1)
else:
cpt_con1 = Concatenate(axis=1)([c_out1, p_out1, t_out1])
if kernel1:
cpt = conv_unit1(pre_F * 3, conv_F, drop, H, W)(cpt_con1)
else:
cpt = conv_unit0(pre_F * 3, conv_F, drop, H, W)(cpt_con1)
if is_plus:
if is_plus_efficient:
for i in range(R_N):
cpt = Res_plus_E('Res_plus_' + str(i + 1), conv_F, plus, rate,
drop, H, W)(cpt)
else:
for i in range(R_N):
cpt = Res_plus('Res_plus_' + str(i + 1), conv_F, plus, rate,
drop, H, W)(cpt)
else:
for i in range(R_N):
cpt = Res_normal('Res_normal_' + str(i + 1), conv_F, drop, H,
W)(cpt)
cpt_conv2 = Activation('relu')(cpt)
cpt_out2 = BatchNormalization()(cpt_conv2)
cpt_conv1 = Dropout(drop)(cpt_out2)
cpt_conv1 = Conv2D(filters=channel, kernel_size=(1, 1),
padding="same")(cpt_conv1)
cpt_out1 = Activation('tanh')(cpt_conv1)
if is_pt:
DeepSTN_model = Model(inputs=[cpt_input, poi_in, time_in],
outputs=cpt_out1)
else:
DeepSTN_model = Model(inputs=cpt_input, outputs=cpt_out1)
DeepSTN_model.compile(loss='mse', optimizer='Adam', metrics=['mae'])
DeepSTN_model.summary()
return DeepSTN_model
def build_model(img_shape: Tuple[int, int, int], num_classes: int,
optimizer: tf.keras.optimizers.Optimizer, learning_rate: float,
filter_block1: int, kernel_size_block1: int,
filter_block2: int, kernel_size_block2: int,
filter_block3: int, kernel_size_block3: int,
dense_layer_size: int,
kernel_initializer: tf.keras.initializers.Initializer,
activation_cls: tf.keras.layers.Activation,
dropout_rate: float, use_batch_normalization: bool,
use_dense: bool, use_global_pooling: bool) -> Model:
input_img = Input(shape=img_shape)
x = Conv2D(filters=filter_block1,
kernel_size=kernel_size_block1,
padding="same",
kernel_initializer=kernel_initializer,
name="heatmap1")(input_img)
if use_batch_normalization:
x = BatchNormalization()(x)
x = activation_cls(x)
x = Conv2D(filters=filter_block1,
kernel_size=kernel_size_block1,
padding="same",
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
if dropout_rate:
x = Dropout(rate=dropout_rate)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
x = Conv2D(filters=filter_block2,
kernel_size=kernel_size_block2,
padding="same",
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
x = activation_cls(x)
x = Conv2D(filters=filter_block2,
kernel_size=kernel_size_block2,
padding="same",
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
if dropout_rate:
x = Dropout(rate=dropout_rate)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
x = Conv2D(filters=filter_block3,
kernel_size=kernel_size_block3,
padding="same",
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
x = activation_cls(x)
x = Conv2D(filters=filter_block3,
kernel_size=kernel_size_block3,
padding="same",
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
if dropout_rate:
x = Dropout(rate=dropout_rate)(x)
x = activation_cls(x)
x = MaxPool2D()(x)
if use_global_pooling:
x = GlobalAveragePooling2D()(x)
else:
x = Flatten()(x)
if use_dense:
x = Dense(units=dense_layer_size,
kernel_initializer=kernel_initializer)(x)
if use_batch_normalization:
x = BatchNormalization()(x)
x = activation_cls(x)
x = Dense(units=num_classes, kernel_initializer=kernel_initializer)(x)
y_pred = Activation("softmax")(x)
model = Model(inputs=[input_img], outputs=[y_pred])
opt = optimizer(learning_rate=learning_rate)
model.compile(loss="categorical_crossentropy",
optimizer=opt,
metrics=["accuracy"])
return model
input_shape = (32, 32, 3)
# downsampling to get image to same input size as model
img_array = resize(img_array, input_shape, anti_aliasing=True)
img_array = img_array.reshape(
(1, img_array.shape[0], img_array.shape[1], img_array.shape[2]))
# the model is called modelo_ar
activator = 'relu'
optimizer = 'adam'
modelo_ar = Sequential()
modelo_ar.add(
Conv2D(32, (3, 3),
activation=activator,
padding='same',
input_shape=input_shape))
modelo_ar.add(Conv2D(32, (3, 3), activation=activator, padding='same'))
modelo_ar.add(Conv2D(32, (3, 3), activation=activator, padding='same'))
modelo_ar.add(MaxPooling2D((2, 2)))
modelo_ar.add(Dropout(0.2))
modelo_ar.add(Conv2D(64, (3, 3), activation=activator, padding='same'))
modelo_ar.add(Conv2D(64, (3, 3), activation=activator, padding='same'))
modelo_ar.add(Conv2D(64, (3, 3), activation=activator, padding='same'))
modelo_ar.add(MaxPooling2D((2, 2)))
modelo_ar.add(Dropout(0.2))
modelo_ar.add(Conv2D(128, (3, 3), activation=activator, padding='same'))
modelo_ar.add(Conv2D(128, (3, 3), activation=activator, padding='same'))
modelo_ar.add(MaxPooling2D((2, 2)))
modelo_ar.add(Flatten())
modelo_ar.add(Dense(128, activation=activator))
def load_vgg19(self):
"""
Returns: Create a Model Template of VGG19 with hyperspectral input shape (as defined in init)
"""
## adapted from https://stackoverflow.com/questions/53251827/pretrained-tensorflow-model-rgb-rgby-channel-extension
## set up vgg19 template with hyperspectral input vector
x = Conv2D(64, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block1_conv1')(self.hs_inputs)
x = Conv2D(64, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block1_conv2')(x)
x = MaxPooling2D(pool_size=(2, 2), name='block1_pool')(x)
# block 2
x = Conv2D(128, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block2_conv1')(x)
x = Conv2D(128, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block2_conv2')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
name='block2_pool')(x)
# block 3
x = Conv2D(256, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block3_conv1')(x)
x = Conv2D(256, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block3_conv2')(x)
x = Conv2D(256, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block3_conv3')(x)
x = Conv2D(256, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block3_conv4')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
name='block3_pool')(x)
# block 4
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block4_conv1')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block4_conv2')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block4_conv3')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block4_conv4')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
name='block4_pool')(x)
# block 5
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block5_conv1')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block5_conv2')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block5_conv3')(x)
x = Conv2D(512, (3, 3),
padding='same',
activation='relu',
kernel_initializer='zeros',
name='block5_conv4')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
name='block5_pool')(x)
x = layers.Flatten(name='flatten')(x)
x = layers.Dense(128, activation='relu')(x)
x = layers.Dense(128, activation='relu')(x)
x = layers.Dense(1)(x)
vgg_template = models.Model(inputs=self.hs_inputs, outputs=x)
if self.channels > 3:
layers_to_modify = ['conv1_conv'
] # Turns out the only layer that changes
# shape due to 4th to 13th channel is the first convolution layer.
else:
layers_to_modify = []
for layer in self.pretrained_model.layers: # pretrained Model and template have the same
# layers, so it doesn't matter which to
# iterate over.
if layer.get_weights(
): # Skip input, pooling and no weights layers
target_layer = vgg_template.get_layer(name=layer.name)
# print(target_layer.get_weights())
if layer.name in layers_to_modify:
kernels = layer.get_weights()[0]
biases = layer.get_weights()[1]
kernels_extra_channel = np.concatenate(
(kernels, kernels[:, :, -3:, :], kernels[:, :, -3:, :],
kernels[:, :, -3:, :], kernels[:, :, -1:, :]),
axis=-2) # For channels_last
target_layer.set_weights([kernels_extra_channel, biases])
else:
target_layer.set_weights(layer.get_weights())
return vgg_template
name_arr = name.split("_")
image_category_list.append(name_arr[1])
y_train = np.asarray(image_category_list)
y_train = utils.to_categorical(y_train, 11)
x_train = load_images_to_np_array(image_file_list)
# print(x_train[0])
# plt.imshow(x_train[(len(x_train) - 4)])
# plt.show()
# losses = {'category_output': 'categorical_crossentropy', 'counter_output': 'mse'}
# strategy = tf.distribute.MirroredStrategy()
# with strategy.scope():
i = Input(shape=x_train[0].shape)
nq_network = Conv2D(35, (3, 3), activation='relu', padding='same')(i)
nq_network = BatchNormalization()(nq_network)
nq_network = MaxPooling2D((2, 2))(nq_network)
nq_network = Conv2D(70, (3, 3), activation='relu', padding='same')(nq_network)
nq_network = BatchNormalization()(nq_network)
nq_network = MaxPooling2D((2, 2))(nq_network)
nq_network = Conv2D(100, (3, 3), activation='relu', padding='same')(nq_network)
nq_network = BatchNormalization()(nq_network)
nq_network = MaxPooling2D((2, 2))(nq_network)
nq_network = Flatten()(nq_network)
nq_network = Dropout(0.2)(nq_network)
nq_network = Dense(50, activation='relu')(nq_network)
nq_network = Dropout(0.1)(nq_network)
nq_network = Dense(11, activation='softmax')(nq_network)
def unet(flags_obj, n_filters=64):
# Contracting Path (encoding)
inputs = Input(flags_obj.input_size)
conv1 = Conv2D(n_filters,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(inputs)
conv1 = Conv2D(n_filters,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(n_filters * 2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool1)
conv2 = Conv2D(n_filters * 2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(n_filters * 4,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool2)
conv3 = Conv2D(n_filters * 4,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(n_filters * 8,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool3)
conv4 = Conv2D(n_filters * 8,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv4)
drop4 = Dropout(0.3)(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(drop4)
conv5 = Conv2D(n_filters * 16,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool4)
conv5 = Conv2D(n_filters * 16,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv5)
drop5 = Dropout(0.3)(conv5)
# Expansive Path (decoding)
up6 = Conv2DTranspose(n_filters * 8, (3, 3), strides=(2, 2),
padding='same')(drop5)
merge6 = concatenate([up6, drop4], axis=3)
conv6 = Conv2D(n_filters * 8,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge6)
conv6 = Conv2D(n_filters * 8,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv6)
up7 = Conv2DTranspose(n_filters * 4, (3, 3), strides=(2, 2),
padding='same')(conv6)
merge7 = concatenate([up7, conv3], axis=3)
conv7 = Conv2D(n_filters * 4,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge7)
conv7 = Conv2D(n_filters * 4,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv7)
up8 = Conv2DTranspose(n_filters * 2, (3, 3), strides=(2, 2),
padding='same')(conv7)
merge8 = concatenate([up8, conv2], axis=3)
conv8 = Conv2D(n_filters * 2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge8)
conv8 = Conv2D(n_filters * 2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv8)
up9 = Conv2DTranspose(n_filters, (3, 3), strides=(2, 2),
padding='same')(conv8)
merge9 = concatenate([up9, conv1], axis=3)
conv9 = Conv2D(n_filters,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge9)
conv9 = Conv2D(n_filters,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv9)
conv9 = Conv2D(2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv9)
conv10 = Conv2D(1, 1, activation='sigmoid')(conv9)
model = tf.keras.Model(inputs=inputs, outputs=conv10)
return model
from tensorflow.keras.layers import Conv2D, MaxPooling2D
Name = "Cats_VS_Dogs_64X2_{}".format(int(time.time()))
tensorboard = TensorBoard(log_dir='logs\{}'.format(Name))
pickle_in = open("x_pickle", "rb")
X = pickle.load(pickle_in)
pickle_in = open("y_pickle", "rb")
y = pickle.load(pickle_in)
X = X / 255.0
X = np.array(X)
y = np.array(y)
model = Sequential()
model.add(Conv2D(64, (3, 3), input_shape=X.shape[1:]))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, (3, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(64))
model.add(Activation("relu"))
model.add(Dense(1))
model.add(Activation("sigmoid"))
model.compile(loss="binary_crossentropy",
import pickle
from tensorflow.keras.layers import Dense,Dropout,Conv2D,MaxPool2D,Flatten
from tensorflow.keras.models import Sequential
import numpy as np
pickle_in = open(" X.pickle","rb")
X=pickle.load(pickle_in)
pickle_in.close()
pickle_in1 = open("Y.pickle","rb")
Y=pickle.load(pickle_in1)
pickle_in.close()
X=X/255.0
model=Sequential()
model.add(Conv2D(64,(3,3),input_shape=X.shape[1:],activation='relu'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPool2D(pool_size=(2,2)))
model.add(Flatten())
model.add(Dense(64,activation='relu'))
model.add(Dense(1,activation='sigmoid'))
model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])
Y=np.array(Y)
model.fit(X,Y,batch_size=32,epochs=3,validation_split=0.3)
def resnet_v1_eembc(input_shape=[32, 32, 3],
num_classes=10,
num_filters=[16, 32, 64],
kernel_sizes=[3, 1],
strides=[1, 2],
l1p=1e-4,
l2p=0):
# Input layer, change kernel size to 7x7 and strides to 2 for an official resnet
inputs = Input(shape=input_shape)
x = Conv2D(num_filters[0],
kernel_size=kernel_sizes[0],
strides=strides[0],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(inputs)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# First stack
# Weight layers
y = Conv2D(num_filters[0],
kernel_size=kernel_sizes[0],
strides=strides[0],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(x)
y = BatchNormalization()(y)
y = Activation('relu')(y)
y = Conv2D(num_filters[0],
kernel_size=kernel_sizes[0],
strides=strides[0],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(y)
y = BatchNormalization()(y)
# Overall residual, connect weight layer and identity paths
x = Add()([x, y])
x = Activation('relu')(x)
# Second stack
# Weight layers
y = Conv2D(num_filters[1],
kernel_size=kernel_sizes[0],
strides=strides[1],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(x)
y = BatchNormalization()(y)
y = Activation('relu')(y)
y = Conv2D(num_filters[1],
kernel_size=kernel_sizes[0],
strides=strides[0],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(y)
y = BatchNormalization()(y)
# Adjust for change in dimension due to stride in identity
x = Conv2D(num_filters[1],
kernel_size=kernel_sizes[1],
strides=strides[1],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(x)
# Overall residual, connect weight layer and identity paths
x = Add()([x, y])
x = Activation('relu')(x)
# Third stack
# Weight layers
y = Conv2D(num_filters[2],
kernel_size=kernel_sizes[0],
strides=strides[1],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(x)
y = BatchNormalization()(y)
y = Activation('relu')(y)
y = Conv2D(num_filters[2],
kernel_size=kernel_sizes[0],
strides=strides[0],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(y)
y = BatchNormalization()(y)
# Adjust for change in dimension due to stride in identity
x = Conv2D(num_filters[2],
kernel_size=kernel_sizes[1],
strides=strides[1],
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l1_l2(l1=l1p, l2=l2p))(x)
# Overall residual, connect weight layer and identity paths
x = Add()([x, y])
x = Activation('relu')(x)
# Final classification layer.
pool_size = int(np.amin(x.shape[1:3]))
x = AveragePooling2D(pool_size=pool_size)(x)
y = Flatten()(x)
outputs = Dense(num_classes,
activation='softmax',
kernel_initializer='he_normal')(y)
# Instantiate model.
model = Model(inputs=inputs, outputs=outputs)
return model
def mini_XCEPTION(input_shape, num_classes, l2_regularization=0.01):
regularization = l2(l2_regularization)
# base
img_input = Input(input_shape)
x = Conv2D(8, (3, 3),
strides=(1, 1),
kernel_regularizer=regularization,
use_bias=False)(img_input)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Conv2D(8, (3, 3),
strides=(1, 1),
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# module 1
residual = Conv2D(16, (1, 1),
strides=(2, 2),
padding='same',
use_bias=False)(x)
residual = BatchNormalization()(residual)
x = SeparableConv2D(16, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = SeparableConv2D(16, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
x = layers.add([x, residual])
# module 2
residual = Conv2D(32, (1, 1),
strides=(2, 2),
padding='same',
use_bias=False)(x)
residual = BatchNormalization()(residual)
x = SeparableConv2D(32, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = SeparableConv2D(32, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
x = layers.add([x, residual])
# module 3
residual = Conv2D(64, (1, 1),
strides=(2, 2),
padding='same',
use_bias=False)(x)
residual = BatchNormalization()(residual)
x = SeparableConv2D(64, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = SeparableConv2D(64, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
x = layers.add([x, residual])
# module 4
residual = Conv2D(128, (1, 1),
strides=(2, 2),
padding='same',
use_bias=False)(x)
residual = BatchNormalization()(residual)
x = SeparableConv2D(128, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = SeparableConv2D(128, (3, 3),
padding='same',
kernel_regularizer=regularization,
use_bias=False)(x)
x = BatchNormalization()(x)
x = MaxPooling2D((3, 3), strides=(2, 2), padding='same')(x)
x = layers.add([x, residual])
x = Conv2D(
num_classes,
(3, 3),
# kernel_regularizer=regularization,
padding='same')(x)
x = GlobalAveragePooling2D()(x)
output = Dense(30)(x)
model = Model(img_input, output)
return model
def build_net(self, dev):
with tf.variable_scope('a3c') and tf.device(dev):
screenInput = Input(
shape=(U.screen_channel(), self.ssize, self.ssize),
name='screenInput',
)
permutedScreenInput = Permute((2,3,1))(screenInput)
conv1 = Conv2D(16, kernel_size=5, strides=(1,1), padding='same',name='conv1')(permutedScreenInput)
conv2 = Conv2D(32, kernel_size=3, strides=(1,1), padding='same',name='conv2')(conv1)
infoInput = Input(
shape=(self.isize,),
name='infoInput',
)
customInput = Input(
shape=(self.custom_input_size,),
name='customInput',
)
nonSpatialInput = Concatenate(name='nonSpatialInputConcat')([infoInput, customInput])
broadcasted = Lambda(self.broadcast,name='broadcasting')(nonSpatialInput)
combinedSpatialNonSpatial = Concatenate(name='combinedConcat')([broadcasted, conv2])
conv3 = Conv2D(1, kernel_size=1, strides=(1,1), padding='same',name='conv3')(combinedSpatialNonSpatial)
flatConv3 = Flatten(name='flatConv3')(conv3)
lstmInput = Lambda(self.expand_dims, name='lstmInput')(flatConv3)
self.NUM_LSTM = 100
hStateInput = Input(
shape=(self.NUM_LSTM,),
name='hStateInput'
)
cStateInput = Input(
shape=(self.NUM_LSTM,),
name='cStateInput'
)
lstm, hStates, cStates = LSTM(self.NUM_LSTM, return_state=True)(lstmInput, initial_state=[hStateInput, cStateInput])
fc1 = Dense(256, activation='relu',name='dense1')(lstm)
fc2 = Dense(1, activation='linear',name='fc2')(fc1)
value = Lambda(self.Squeeze,name='value')(fc2)
policy = Dense(self.isize, activation='softmax',name='policy')(fc1)
broadcastLstm = Lambda(self.broadcast, name='breadcastLstm')(lstm)
spatialLstm = Concatenate(name='spatialLstm')([conv3, broadcastLstm])
conv4 = Conv2D(1,kernel_size=1, strides=(1,1), padding='same',name='conv4')(spatialLstm)
flatConv4 = Flatten(name='flattenedConv3')(conv4)
spatialPolicy = Softmax(name='spatialPolicy')(flatConv4)
conv5 = Conv2D(1, kernel_size=1, strides=(1,1), padding='same',name='conv5')(spatialLstm)
flatConv5 = Flatten(name='flattenedConv5')(conv5)
bestRoach = Softmax(name='bestRoach')(flatConv5)
self.model = Model(
inputs=[screenInput, infoInput, customInput, hStateInput, cStateInput],
outputs=[value, policy, spatialPolicy, hStates, cStates, bestRoach]
)
self.model._make_predict_function()
def __init__(self, latent_code_garms_sz, latent_code_betas_sz, name=None):
super(SingleImageNet, self).__init__(name=name)
# Define layers
if IMG_SIZE >= 1000:
self.conv1 = Conv2D(8, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv1_1 = Conv2D(16, (3, 3),
strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
else:
self.conv1 = Conv2D(8, (3, 3),
activation='relu',
kernel_initializer='he_normal',
padding='same')
self.conv1_1 = Conv2D(16, (3, 3),
activation='relu',
kernel_initializer='he_normal',
padding='same')
self.conv2 = Conv2D(
16,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv2_1 = Conv2D(
32,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv3 = Conv2D(
16,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv3_1 = Conv2D(
32,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
sz = 16
self.conv4 = Conv2D(
sz,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv4_1 = Conv2D(
sz,
(3, 3), # strides=(2, 2),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv5 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv5_1 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv6 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv6_1 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv7 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.conv7_1 = Conv2D(sz, (3, 3),
kernel_initializer='he_normal',
padding='same',
activation='relu')
self.split_shape = Lambda(lambda z: z[..., :int(sz / 2)])
self.split_garms = Lambda(lambda z: z[..., int(sz / 2):])
self.flatten = Flatten()
self.dg = Dense(latent_code_garms_sz,
kernel_initializer='he_normal',
activation='relu')
self.db = Dense(latent_code_betas_sz,
kernel_initializer='he_normal',
activation='relu')
self.dg2 = Dense(int(latent_code_garms_sz / 2),
kernel_initializer='he_normal',
activation='relu')
self.db2 = Dense(latent_code_betas_sz,
kernel_initializer='he_normal',
activation='relu')
self.dg3 = Dense(int(latent_code_garms_sz / 2),
kernel_initializer='he_normal',
activation='relu')
self.db3 = Dense(latent_code_betas_sz,
kernel_initializer='he_normal',
activation='relu')
self.latent_code_garms = Dense(int(latent_code_garms_sz / 2),
kernel_initializer='he_normal',
name='latent_garms',
activation='relu')
self.latent_code_shape = Dense(latent_code_betas_sz,
kernel_initializer='he_normal',
name='latent_shape',
activation='relu')
self.concat = Concatenate()
def fully_convolutional_resnet50(
input_shape,
num_classes=1000,
pretrained_resnet=True,
use_bias=True,
):
# init input layer
img_input = Input(shape=input_shape)
# define basic model pipeline
x = ZeroPadding2D(padding=((3, 3), (3, 3)), name="conv1_pad")(img_input)
x = Conv2D(64, 7, strides=2, use_bias=use_bias, name="conv1_conv")(x)
x = BatchNormalization(axis=3, epsilon=1.001e-5, name="conv1_bn")(x)
x = Activation("relu", name="conv1_relu")(x)
x = ZeroPadding2D(padding=((1, 1), (1, 1)), name="pool1_pad")(x)
x = MaxPooling2D(3, strides=2, name="pool1_pool")(x)
# the sequence of stacked residual blocks
x = stack1(x, 64, 3, stride1=1, name="conv2")
x = stack1(x, 128, 4, name="conv3")
x = stack1(x, 256, 6, name="conv4")
x = stack1(x, 512, 3, name="conv5")
# add avg pooling layer after feature extraction layers
x = AveragePooling2D(pool_size=7)(x)
# add final convolutional layer
conv_layer_final = Conv2D(
filters=num_classes,
kernel_size=1,
use_bias=use_bias,
name="last_conv",
)(x)
# configure fully convolutional ResNet50 model
model = training.Model(img_input, x)
# load model weights
if pretrained_resnet:
model_name = "resnet50"
# configure full file name
file_name = model_name + "_weights_tf_dim_ordering_tf_kernels_notop.h5"
# get the file hash from TF WEIGHTS_HASHES
file_hash = WEIGHTS_HASHES[model_name][1]
weights_path = data_utils.get_file(
file_name,
BASE_WEIGHTS_PATH + file_name,
cache_subdir="models",
file_hash=file_hash,
)
model.load_weights(weights_path)
# form final model
model = training.Model(inputs=model.input, outputs=[conv_layer_final])
if pretrained_resnet:
# get model with the dense layer for further FC weights extraction
resnet50_extractor = ResNet50(
include_top=True,
weights="imagenet",
classes=num_classes,
)
# set ResNet50 FC-layer weights to final convolutional layer
set_conv_weights(model=model, feature_extractor=resnet50_extractor)
return model
image = cv2.imread(image_file)
rgbimage = cv2.cvtColor(image, cv2.COLOR_RGBA2RGB)
imgresize = resize_to_fit(rgbimage, 120, 120)
images.append(imgresize / 255.0)
images = np.array(images)
data = pd.read_csv('recap.csv')
print(images.shape)
labels = np.array(data)
print(labels.shape)
images, labels = shuffle(images, labels)
X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(images, labels, test_size=0.1)
model = Sequential()
model.add(Conv2D(16, (3, 3), activation='relu',))
model.add(BatchNormalization())
model.add(MaxPool2D(2, 2))
model.add(Dropout(.2))
model.add(Conv2D(32, (3, 3), activation='relu',))
model.add(BatchNormalization())
model.add(MaxPool2D(2, 2))
model.add(Dropout(.3))
model.add(Conv2D(64, (3, 3), activation='relu',))
model.add(BatchNormalization())
model.add(MaxPool2D(2, 2))
model.add(Dropout(.4))
model.add(Conv2D(128, (3, 3), activation='relu', ))
dropout_rate = 0.25
save_dir = os.path.join(os.getcwd(), "..", "saved_models")
save_dir = os.path.abspath(save_dir)
model_name = "dropout_rate_025.h5"
print(model_name)
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train = x_train / 255
x_test = x_test / 255
y_train = keras.utils.to_categorical(y_train, classes)
y_test = keras.utils.to_categorical(y_test, classes)
model = Sequential()
model.add(Conv2D(32, (3, 3), padding="SAME", input_shape=x_train.shape[1:]))
model.add(Activation("relu"))
model.add(Conv2D(32, (3, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(dropout_rate))
model.add(Conv2D(64, (3, 3), padding="SAME"))
model.add(Activation("relu"))
model.add(Conv2D(64, (3, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(dropout_rate))
model.add(Flatten())
model.add(Dense(512))
def bluche(input_size, d_model):
"""
Gated Convolucional Recurrent Neural Network by Bluche et al.
Reference:
Bluche, T., Messina, R.:
Gated convolutional recurrent neural networks for multilingual handwriting recognition.
In: Document Analysis and Recognition (ICDAR), 2017
14th IAPR International Conference on, vol. 1, pp. 646–651, 2017.
URL: https://ieeexplore.ieee.org/document/8270042
"""
input_data = Input(name="input", shape=input_size)
cnn = Reshape((input_size[0] // 2, input_size[1] // 2,
input_size[2] * 4))(input_data)
cnn = Conv2D(filters=8,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
activation="tanh")(cnn)
cnn = Conv2D(filters=16,
kernel_size=(2, 4),
strides=(2, 4),
padding="same",
activation="tanh")(cnn)
cnn = GatedConv2D(filters=16,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = Conv2D(filters=32,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
activation="tanh")(cnn)
cnn = GatedConv2D(filters=32,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = Conv2D(filters=64,
kernel_size=(2, 4),
strides=(2, 4),
padding="same",
activation="tanh")(cnn)
cnn = GatedConv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same")(cnn)
cnn = Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
activation="tanh")(cnn)
cnn = MaxPooling2D(pool_size=(1, 4), strides=(1, 4), padding="valid")(cnn)
shape = cnn.get_shape()
blstm = Reshape((shape[1], shape[2] * shape[3]))(cnn)
blstm = Bidirectional(LSTM(units=128, return_sequences=True))(blstm)
blstm = Dense(units=128, activation="tanh")(blstm)
blstm = Bidirectional(LSTM(units=128, return_sequences=True))(blstm)
output_data = Dense(units=d_model, activation="softmax")(blstm)
return (input_data, output_data)
norm = BatchNormalization()(gru)
norm = Dropout(0.2)(norm)
reshape = Reshape((SEQ_LENGTH, norm.shape[-1] * frames_per_annotation))(norm)
speech_features = TimeDistributed(Dense(feature_vector_size,
activation='relu'))(reshape)
## conv3d network for video model
img_x = 48
img_y = 48
ch_n = 1
video_input = Input(shape=(SEQ_LENGTH, img_x, img_y, ch_n), name='video_input')
layer = TimeDistributed(
Conv2D(32, kernel_size=(3, 3), activation='relu',
padding='same'))(video_input)
layer = TimeDistributed(BatchNormalization())(layer)
layer = TimeDistributed(
Conv2D(32, kernel_size=(3, 3), activation='relu', padding='same'))(layer)
layer = TimeDistributed(BatchNormalization())(layer)
layer = TimeDistributed(MaxPooling2D(pool_size=(2, 2), padding='same'))(layer)
layer = TimeDistributed(
Conv2D(64, kernel_size=(3, 3), activation='relu', padding='same'))(layer)
layer = TimeDistributed(BatchNormalization())(layer)
layer = TimeDistributed(
Conv2D(64, kernel_size=(3, 3), activation='relu', padding='same'))(layer)
layer = TimeDistributed(BatchNormalization())(layer)
layer = TimeDistributed(MaxPooling2D(pool_size=(2, 2), padding='same'))(layer)
layer = TimeDistributed(Dropout(0.2))(layer)
def flor(input_size, d_model):
"""
Gated Convolucional Recurrent Neural Network by Flor et al.
"""
input_data = Input(name="input", shape=input_size)
cnn = Conv2D(filters=16,
kernel_size=(3, 3),
strides=(1, 2),
padding="same",
kernel_initializer="he_uniform")(input_data)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
cnn = FullGatedConv2D(filters=16, kernel_size=(3, 3), padding="same")(cnn)
cnn = Conv2D(filters=32,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
kernel_initializer="he_uniform")(cnn)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
cnn = FullGatedConv2D(filters=32, kernel_size=(3, 3), padding="same")(cnn)
cnn = Conv2D(filters=40,
kernel_size=(2, 4),
strides=(2, 4),
padding="same",
kernel_initializer="he_uniform")(cnn)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
cnn = FullGatedConv2D(filters=40,
kernel_size=(3, 3),
padding="same",
kernel_constraint=MaxNorm(4, [0, 1, 2]))(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=48,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
kernel_initializer="he_uniform")(cnn)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
cnn = FullGatedConv2D(filters=48,
kernel_size=(3, 3),
padding="same",
kernel_constraint=MaxNorm(4, [0, 1, 2]))(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=56,
kernel_size=(2, 4),
strides=(2, 4),
padding="same",
kernel_initializer="he_uniform")(cnn)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
cnn = FullGatedConv2D(filters=56,
kernel_size=(3, 3),
padding="same",
kernel_constraint=MaxNorm(4, [0, 1, 2]))(cnn)
cnn = Dropout(rate=0.2)(cnn)
cnn = Conv2D(filters=64,
kernel_size=(3, 3),
strides=(1, 1),
padding="same",
kernel_initializer="he_uniform")(cnn)
cnn = PReLU(shared_axes=[1, 2])(cnn)
cnn = BatchNormalization(renorm=True)(cnn)
shape = cnn.get_shape()
bgru = Reshape((shape[1], shape[2] * shape[3]))(cnn)
bgru = Bidirectional(GRU(units=128, return_sequences=True,
dropout=0.5))(bgru)
bgru = Dense(units=256)(bgru)
bgru = Bidirectional(GRU(units=128, return_sequences=True,
dropout=0.5))(bgru)
output_data = Dense(units=d_model, activation="softmax")(bgru)
return (input_data, output_data)
def build_model(image_size,
n_classes,
mode='training',
l2_regularization=0.0,
min_scale=0.1,
max_scale=0.9,
scales=None,
aspect_ratios_global=[0.5, 1.0, 2.0],
aspect_ratios_per_layer=None,
two_boxes_for_ar1=True,
steps=None,
offsets=None,
clip_boxes=False,
variances=[1.0, 1.0, 1.0, 1.0],
coords='centroids',
normalize_coords=False,
subtract_mean=None,
divide_by_stddev=None,
swap_channels=False,
confidence_thresh=0.01,
iou_threshold=0.45,
top_k=200,
nms_max_output_size=400,
return_predictor_sizes=False):
'''
Build a Keras model with SSD architecture, see references.
The model consists of convolutional feature layers and a number of convolutional
predictor layers that take their input from different feature layers.
The model is fully convolutional.
The implementation found here is a smaller version of the original architecture
used in the paper (where the base network consists of a modified VGG-16 extended
by a few convolutional feature layers), but of course it could easily be changed to
an arbitrarily large SSD architecture by following the general design pattern used here.
This implementation has 7 convolutional layers and 4 convolutional predictor
layers that take their input from layers 4, 5, 6, and 7, respectively.
Most of the arguments that this function takes are only needed for the anchor
box layers. In case you're training the network, the parameters passed here must
be the same as the ones used to set up `SSDBoxEncoder`. In case you're loading
trained weights, the parameters passed here must be the same as the ones used
to produce the trained weights.
Some of these arguments are explained in more detail in the documentation of the
`SSDBoxEncoder` class.
Note: Requires Keras v2.0 or later. Training currently works only with the
TensorFlow backend (v1.0 or later).
Arguments:
image_size (tuple): The input image size in the format `(height, width, channels)`.
n_classes (int): The number of positive classes, e.g. 20 for Pascal VOC, 80 for MS COCO.
mode (str, optional): One of 'training', 'inference' and 'inference_fast'. In 'training' mode,
the model outputs the raw prediction tensor, while in 'inference' and 'inference_fast' modes,
the raw predictions are decoded into absolute coordinates and filtered via confidence thresholding,
non-maximum suppression, and top-k filtering. The difference between latter two modes is that
'inference' follows the exact procedure of the original Caffe implementation, while
'inference_fast' uses a faster prediction decoding procedure.
l2_regularization (float, optional): The L2-regularization rate. Applies to all convolutional layers.
min_scale (float, optional): The smallest scaling factor for the size of the anchor boxes as a fraction
of the shorter side of the input images.
max_scale (float, optional): The largest scaling factor for the size of the anchor boxes as a fraction
of the shorter side of the input images. All scaling factors between the smallest and the
largest will be linearly interpolated. Note that the second to last of the linearly interpolated
scaling factors will actually be the scaling factor for the last predictor layer, while the last
scaling factor is used for the second box for aspect ratio 1 in the last predictor layer
if `two_boxes_for_ar1` is `True`.
scales (list, optional): A list of floats containing scaling factors per convolutional predictor layer.
This list must be one element longer than the number of predictor layers. The first `k` elements are the
scaling factors for the `k` predictor layers, while the last element is used for the second box
for aspect ratio 1 in the last predictor layer if `two_boxes_for_ar1` is `True`. This additional
last scaling factor must be passed either way, even if it is not being used. If a list is passed,
this argument overrides `min_scale` and `max_scale`. All scaling factors must be greater than zero.
aspect_ratios_global (list, optional): The list of aspect ratios for which anchor boxes are to be
generated. This list is valid for all predictor layers. The original implementation uses more aspect ratios
for some predictor layers and fewer for others. If you want to do that, too, then use the next argument instead.
aspect_ratios_per_layer (list, optional): A list containing one aspect ratio list for each predictor layer.
This allows you to set the aspect ratios for each predictor layer individually. If a list is passed,
it overrides `aspect_ratios_global`.
two_boxes_for_ar1 (bool, optional): Only relevant for aspect ratio lists that contain 1. Will be ignored otherwise.
If `True`, two anchor boxes will be generated for aspect ratio 1. The first will be generated
using the scaling factor for the respective layer, the second one will be generated using
geometric mean of said scaling factor and next bigger scaling factor.
steps (list, optional): `None` or a list with as many elements as there are predictor layers. The elements can be
either ints/floats or tuples of two ints/floats. These numbers represent for each predictor layer how many
pixels apart the anchor box center points should be vertically and horizontally along the spatial grid over
the image. If the list contains ints/floats, then that value will be used for both spatial dimensions.
If the list contains tuples of two ints/floats, then they represent `(step_height, step_width)`.
If no steps are provided, then they will be computed such that the anchor box center points will form an
equidistant grid within the image dimensions.
offsets (list, optional): `None` or a list with as many elements as there are predictor layers. The elements can be
either floats or tuples of two floats. These numbers represent for each predictor layer how many
pixels from the top and left boarders of the image the top-most and left-most anchor box center points should be
as a fraction of `steps`. The last bit is important: The offsets are not absolute pixel values, but fractions
of the step size specified in the `steps` argument. If the list contains floats, then that value will
be used for both spatial dimensions. If the list contains tuples of two floats, then they represent
`(vertical_offset, horizontal_offset)`. If no offsets are provided, then they will default to 0.5 of the step size,
which is also the recommended setting.
clip_boxes (bool, optional): If `True`, clips the anchor box coordinates to stay within image boundaries.
variances (list, optional): A list of 4 floats >0. The anchor box offset for each coordinate will be divided by
its respective variance value.
coords (str, optional): The box coordinate format to be used internally by the model (i.e. this is not the input format
of the ground truth labels). Can be either 'centroids' for the format `(cx, cy, w, h)` (box center coordinates, width,
and height), 'minmax' for the format `(xmin, xmax, ymin, ymax)`, or 'corners' for the format `(xmin, ymin, xmax, ymax)`.
normalize_coords (bool, optional): Set to `True` if the model is supposed to use relative instead of absolute coordinates,
i.e. if the model predicts box coordinates within [0,1] instead of absolute coordinates.
subtract_mean (array-like, optional): `None` or an array-like object of integers or floating point values
of any shape that is broadcast-compatible with the image shape. The elements of this array will be
subtracted from the image pixel intensity values. For example, pass a list of three integers
to perform per-channel mean normalization for color images.
divide_by_stddev (array-like, optional): `None` or an array-like object of non-zero integers or
floating point values of any shape that is broadcast-compatible with the image shape. The image pixel
intensity values will be divided by the elements of this array. For example, pass a list
of three integers to perform per-channel standard deviation normalization for color images.
swap_channels (list, optional): Either `False` or a list of integers representing the desired order in which the input
image channels should be swapped.
confidence_thresh (float, optional): A float in [0,1), the minimum classification confidence in a specific
positive class in order to be considered for the non-maximum suppression stage for the respective class.
A lower value will result in a larger part of the selection process being done by the non-maximum suppression
stage, while a larger value will result in a larger part of the selection process happening in the confidence
thresholding stage.
iou_threshold (float, optional): A float in [0,1]. All boxes that have a Jaccard similarity of greater than `iou_threshold`
with a locally maximal box will be removed from the set of predictions for a given class, where 'maximal' refers
to the box's confidence score.
top_k (int, optional): The number of highest scoring predictions to be kept for each batch item after the
non-maximum suppression stage.
nms_max_output_size (int, optional): The maximal number of predictions that will be left over after the NMS stage.
return_predictor_sizes (bool, optional): If `True`, this function not only returns the model, but also
a list containing the spatial dimensions of the predictor layers. This isn't strictly necessary since
you can always get their sizes easily via the Keras API, but it's convenient and less error-prone
to get them this way. They are only relevant for training anyway (SSDBoxEncoder needs to know the
spatial dimensions of the predictor layers), for inference you don't need them.
Returns:
model: The Keras SSD model.
predictor_sizes (optional): A Numpy array containing the `(height, width)` portion
of the output tensor shape for each convolutional predictor layer. During
training, the generator function needs this in order to transform
the ground truth labels into tensors of identical structure as the
output tensors of the model, which is in turn needed for the cost
function.
References:
https://arxiv.org/abs/1512.02325v5
'''
n_predictor_layers = 4 # The number of predictor conv layers in the network
n_classes += 1 # Account for the background class.
l2_reg = l2_regularization # Make the internal name shorter.
img_height, img_width, img_channels = image_size[0], image_size[1], image_size[2]
############################################################################
# Get a few exceptions out of the way.
############################################################################
if aspect_ratios_global is None and aspect_ratios_per_layer is None:
raise ValueError("`aspect_ratios_global` and `aspect_ratios_per_layer` cannot both be None. At least one needs to be specified.")
if aspect_ratios_per_layer:
if len(aspect_ratios_per_layer) != n_predictor_layers:
raise ValueError("It must be either aspect_ratios_per_layer is None or len(aspect_ratios_per_layer) == {}, but len(aspect_ratios_per_layer) == {}.".format(n_predictor_layers, len(aspect_ratios_per_layer)))
if (min_scale is None or max_scale is None) and scales is None:
raise ValueError("Either `min_scale` and `max_scale` or `scales` need to be specified.")
if scales:
if len(scales) != n_predictor_layers+1:
raise ValueError("It must be either scales is None or len(scales) == {}, but len(scales) == {}.".format(n_predictor_layers+1, len(scales)))
else: # If no explicit list of scaling factors was passed, compute the list of scaling factors from `min_scale` and `max_scale`
scales = np.linspace(min_scale, max_scale, n_predictor_layers+1)
if len(variances) != 4: # We need one variance value for each of the four box coordinates
raise ValueError("4 variance values must be pased, but {} values were received.".format(len(variances)))
variances = np.array(variances)
if np.any(variances <= 0):
raise ValueError("All variances must be >0, but the variances given are {}".format(variances))
if (not (steps is None)) and (len(steps) != n_predictor_layers):
raise ValueError("You must provide at least one step value per predictor layer.")
if (not (offsets is None)) and (len(offsets) != n_predictor_layers):
raise ValueError("You must provide at least one offset value per predictor layer.")
############################################################################
# Compute the anchor box parameters.
############################################################################
# Set the aspect ratios for each predictor layer. These are only needed for the anchor box layers.
if aspect_ratios_per_layer:
aspect_ratios = aspect_ratios_per_layer
else:
aspect_ratios = [aspect_ratios_global] * n_predictor_layers
# Compute the number of boxes to be predicted per cell for each predictor layer.
# We need this so that we know how many channels the predictor layers need to have.
if aspect_ratios_per_layer:
n_boxes = []
for ar in aspect_ratios_per_layer:
if (1 in ar) & two_boxes_for_ar1:
n_boxes.append(len(ar) + 1) # +1 for the second box for aspect ratio 1
else:
n_boxes.append(len(ar))
else: # If only a global aspect ratio list was passed, then the number of boxes is the same for each predictor layer
if (1 in aspect_ratios_global) & two_boxes_for_ar1:
n_boxes = len(aspect_ratios_global) + 1
else:
n_boxes = len(aspect_ratios_global)
n_boxes = [n_boxes] * n_predictor_layers
if steps is None:
steps = [None] * n_predictor_layers
if offsets is None:
offsets = [None] * n_predictor_layers
############################################################################
# Define functions for the Lambda layers below.
############################################################################
def identity_layer(tensor):
return tensor
def input_mean_normalization(tensor):
return tensor - np.array(subtract_mean)
def input_stddev_normalization(tensor):
return tensor / np.array(divide_by_stddev)
#def input_channel_swap(tensor):
# if len(swap_channels) == 3:
# return K.stack([tensor[...,swap_channels[0]], tensor[...,swap_channels[1]], tensor[...,swap_channels[2]]], axis=-1)
# elif len(swap_channels) == 4:
# return K.stack([tensor[...,swap_channels[0]], tensor[...,swap_channels[1]], tensor[...,swap_channels[2]], tensor[...,swap_channels[3]]], axis=-1)
############################################################################
# Build the network.
############################################################################
x = Input(shape=(img_height, img_width, img_channels))
# The following identity layer is only needed so that the subsequent lambda layers can be optional.
x1 = Lambda(identity_layer, output_shape=(img_height, img_width, img_channels), name='identity_layer')(x)
if not (subtract_mean is None):
x1 = Lambda(input_mean_normalization, output_shape=(img_height, img_width, img_channels), name='input_mean_normalization')(x1)
if not (divide_by_stddev is None):
x1 = Lambda(input_stddev_normalization, output_shape=(img_height, img_width, img_channels), name='input_stddev_normalization')(x1)
#if swap_channels: #REMOVED FOR TFLITE
# x1 = Lambda(input_channel_swap, output_shape=(img_height, img_width, img_channels), name='input_channel_swap')(x1)
conv1 = Conv2D(32, (5, 5), strides=(2, 2), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv1')(x1)
conv1 = BatchNormalization(axis=3, momentum=0.99, name='bn1')(conv1) # Tensorflow uses filter format [filter_height, filter_width, in_channels, out_channels], hence axis = 3
conv1 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding='same', name='pool1')(conv1)
conv2 = QuantConv2D(48, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv2', **kwargs)(conv1)
conv2 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool2')(conv2)
conv2 = BatchNormalization(axis=3, momentum=0.99, name='bn2')(conv2)
conv3 = QuantConv2D(64, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv3', **kwargs)(conv2)
#conv3 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool3')(conv3)
conv3 = BatchNormalization(axis=3, momentum=0.99, name='bn3')(conv3)
#Conv1,2,3 are for feature extraction and downsampling
#Conv4,5,6,7 are the pre-detection feature maps
conv4 = QuantConv2D(64, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv4', **kwargs)(conv3)
conv4 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool4')(conv4)
conv4 = BatchNormalization(axis=3, momentum=0.99, name='bn4')(conv4)
conv5 = QuantConv2D(48, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv5', **kwargs)(conv4)
conv5 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool5')(conv5)
conv5 = BatchNormalization(axis=3, momentum=0.99, name='bn5')(conv5)
conv6 = QuantConv2D(48, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv6', **kwargs)(conv5)
conv6 = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), name='pool6')(conv6)
conv6 = BatchNormalization(axis=3, momentum=0.99, name='bn6')(conv6)
conv7 = QuantConv2D(32, (3, 3), strides=(1, 1), padding="same", kernel_initializer='glorot_normal', use_bias=False, name='conv7', **kwargs)(conv6)
conv7 = BatchNormalization(axis=3, momentum=0.99, name='bn7')(conv7)
# The next part is to add the convolutional predictor layers on top of the base network
# that we defined above. Note that I use the term "base network" differently than the paper does.
# To me, the base network is everything that is not convolutional predictor layers or anchor
# box layers. In this case we'll have four predictor layers, but of course you could
# easily rewrite this into an arbitrarily deep base network and add an arbitrary number of
# predictor layers on top of the base network by simply following the pattern shown here.
# Build the convolutional predictor layers on top of conv layers 4, 5, 6, and 7.
# We build two predictor layers on top of each of these layers: One for class prediction (classification), one for box coordinate prediction (localization)
# We precidt `n_classes` confidence values for each box, hence the `classes` predictors have depth `n_boxes * n_classes`
# We predict 4 box coordinates for each box, hence the `boxes` predictors have depth `n_boxes * 4`
# Output shape of `classes`: `(batch, height, width, n_boxes * n_classes)`
classes4 = Conv2D(n_boxes[0] * n_classes, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='classes4')(conv4)
classes5 = Conv2D(n_boxes[1] * n_classes, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='classes5')(conv5)
classes6 = Conv2D(n_boxes[2] * n_classes, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='classes6')(conv6)
classes7 = Conv2D(n_boxes[3] * n_classes, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='classes7')(conv7)
# Output shape of `boxes`: `(batch, height, width, n_boxes * 4)`
boxes4 = Conv2D(n_boxes[0] * 4, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='boxes4')(conv4)
boxes5 = Conv2D(n_boxes[1] * 4, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='boxes5')(conv5)
boxes6 = Conv2D(n_boxes[2] * 4, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='boxes6')(conv6)
boxes7 = Conv2D(n_boxes[3] * 4, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='boxes7')(conv7)
# Generate the anchor boxes
# Output shape of `anchors`: `(batch, height, width, n_boxes, 8)`
anchors4 = AnchorBoxes(img_height, img_width, this_scale=scales[0], next_scale=scales[1], aspect_ratios=aspect_ratios[0],
two_boxes_for_ar1=two_boxes_for_ar1, this_steps=steps[0], this_offsets=offsets[0],
clip_boxes=clip_boxes, variances=variances, coords=coords, normalize_coords=normalize_coords, name='anchors4')(boxes4)
anchors5 = AnchorBoxes(img_height, img_width, this_scale=scales[1], next_scale=scales[2], aspect_ratios=aspect_ratios[1],
two_boxes_for_ar1=two_boxes_for_ar1, this_steps=steps[1], this_offsets=offsets[1],
clip_boxes=clip_boxes, variances=variances, coords=coords, normalize_coords=normalize_coords, name='anchors5')(boxes5)
anchors6 = AnchorBoxes(img_height, img_width, this_scale=scales[2], next_scale=scales[3], aspect_ratios=aspect_ratios[2],
two_boxes_for_ar1=two_boxes_for_ar1, this_steps=steps[2], this_offsets=offsets[2],
clip_boxes=clip_boxes, variances=variances, coords=coords, normalize_coords=normalize_coords, name='anchors6')(boxes6)
anchors7 = AnchorBoxes(img_height, img_width, this_scale=scales[3], next_scale=scales[4], aspect_ratios=aspect_ratios[3],
two_boxes_for_ar1=two_boxes_for_ar1, this_steps=steps[3], this_offsets=offsets[3],
clip_boxes=clip_boxes, variances=variances, coords=coords, normalize_coords=normalize_coords, name='anchors7')(boxes7)
# Reshape the class predictions, yielding 3D tensors of shape `(batch, height * width * n_boxes, n_classes)`
# We want the classes isolated in the last axis to perform softmax on them
#Shape inference is different for keras.layers and tf.keras.layers (-1), check documentation
#So I had to manually input the intended shape of the reshape
#cba = classes, boxes, anchors SHAPE: shape will be reused later anyways
cba_4 = classes4.shape[1]*classes4.shape[2]*n_boxes[0]
cba_5 = classes5.shape[1]*classes5.shape[2]*n_boxes[1]
cba_6 = classes6.shape[1]*classes6.shape[2]*n_boxes[2]
cba_7 = classes7.shape[1]*classes7.shape[2]*n_boxes[3]
classes4_reshaped = Reshape((cba_4, n_classes), name='classes4_reshape')(classes4)
classes5_reshaped = Reshape((cba_5, n_classes), name='classes5_reshape')(classes5)
classes6_reshaped = Reshape((cba_6, n_classes), name='classes6_reshape')(classes6)
classes7_reshaped = Reshape((cba_7, n_classes), name='classes7_reshape')(classes7)
# Reshape the box coordinate predictions, yielding 3D tensors of shape `(batch, height * width * n_boxes, 4)`
#shape for classes_reshaped is SAME with boxes EXCEPT for n_classes and anchors so NO NEED TO RECOMPUTE
# We want the four box coordinates isolated in the last axis to compute the smooth L1 loss
boxes4_reshaped = Reshape((cba_4, 4), name='boxes4_reshape')(boxes4)
boxes5_reshaped = Reshape((cba_5, 4), name='boxes5_reshape')(boxes5)
boxes6_reshaped = Reshape((cba_6, 4), name='boxes6_reshape')(boxes6)
boxes7_reshaped = Reshape((cba_7, 4), name='boxes7_reshape')(boxes7)
# Reshape the anchor box tensors, yielding 3D tensors of shape `(batch, height * width * n_boxes, 8)`
anchors4_reshaped = Reshape((cba_4, 8), name='anchors4_reshape')(anchors4)
anchors5_reshaped = Reshape((cba_5, 8), name='anchors5_reshape')(anchors5)
anchors6_reshaped = Reshape((cba_6, 8), name='anchors6_reshape')(anchors6)
anchors7_reshaped = Reshape((cba_7, 8), name='anchors7_reshape')(anchors7)
# Concatenate the predictions from the different layers and the assosciated anchor box tensors
# Axis 0 (batch) and axis 2 (n_classes or 4, respectively) are identical for all layer predictions,
# so we want to concatenate along axis 1
# Output shape of `classes_concat`: (batch, n_boxes_total, n_classes)
classes_concat = Concatenate(axis=1, name='classes_concat')([classes4_reshaped,
classes5_reshaped,
classes6_reshaped,
classes7_reshaped])
# Output shape of `boxes_concat`: (batch, n_boxes_total, 4)
boxes_concat = Concatenate(axis=1, name='boxes_concat')([boxes4_reshaped,
boxes5_reshaped,
boxes6_reshaped,
boxes7_reshaped])
# Output shape of `anchors_concat`: (batch, n_boxes_total, 8)
anchors_concat = Concatenate(axis=1, name='anchors_concat')([anchors4_reshaped,
anchors5_reshaped,
anchors6_reshaped,
anchors7_reshaped])
# The box coordinate predictions will go into the loss function just the way they are,
# but for the class predictions, we'll apply a softmax activation layer first
classes_softmax = Activation('softmax', name='classes_softmax')(classes_concat)
# Concatenate the class and box coordinate predictions and the anchors to one large predictions tensor
# Output shape of `predictions`: (batch, n_boxes_total, n_classes + 4 + 8)
predictions = Concatenate(axis=2, name='predictions')([classes_softmax, boxes_concat, anchors_concat])
if mode == 'training':
model = Model(inputs=x, outputs=predictions)
elif mode == 'inference':
decoded_predictions = DecodeDetections(confidence_thresh=confidence_thresh,
iou_threshold=iou_threshold,
top_k=top_k,
nms_max_output_size=nms_max_output_size,
coords=coords,
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width,
name='decoded_predictions')(predictions)
model = Model(inputs=x, outputs=decoded_predictions)
elif mode == 'inference_fast':
decoded_predictions = DecodeDetectionsFast(confidence_thresh=confidence_thresh,
iou_threshold=iou_threshold,
top_k=top_k,
nms_max_output_size=nms_max_output_size,
coords=coords,
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width,
name='decoded_predictions')(predictions)
model = Model(inputs=x, outputs=decoded_predictions)
else:
raise ValueError("`mode` must be one of 'training', 'inference' or 'inference_fast', but received '{}'.".format(mode))
if return_predictor_sizes:
# The spatial dimensions are the same for the `classes` and `boxes` predictor layers.
predictor_sizes = np.array([classes4._keras_shape[1:3],
classes5._keras_shape[1:3],
classes6._keras_shape[1:3],
classes7._keras_shape[1:3]])
return model, predictor_sizes
else:
return model
def __init__(self):
super(MyModel, self).__init__()
self.conv1 = Conv2D(32, 3, activation='relu')
self.flatten = Flatten()
self.d1 = Dense(128, activation='relu')
self.d2 = Dense(10, activation='softmax')
'''
https://stackoverflow.com/questions/50124158/keras-loss-function-with-additional-dynamic-parameter
'''
from tensorflow.keras.layers import Input, Dense, Conv2D, MaxPool2D, Flatten
from tensorflow.keras.models import Model
from tensorflow.keras.losses import categorical_crossentropy
def sample_loss( y_true, y_pred, is_weight ) :
return is_weight * categorical_crossentropy( y_true, y_pred )
x = Input(shape=(32,32,3), name='image_in')
y_true = Input( shape=(10,), name='y_true' )
is_weight = Input(shape=(1,), name='is_weight')
f = Conv2D(16,(3,3),padding='same')(x)
f = MaxPool2D((2,2),padding='same')(f)
f = Conv2D(32,(3,3),padding='same')(f)
f = MaxPool2D((2,2),padding='same')(f)
f = Conv2D(64,(3,3),padding='same')(f)
f = MaxPool2D((2,2),padding='same')(f)
f = Flatten()(f)
y_pred = Dense(10, activation='softmax', name='y_pred' )(f)
model = Model( inputs=[x, y_true, is_weight], outputs=y_pred, name='train_only' )
model.add_loss( sample_loss( y_true, y_pred, is_weight ) )
model.compile(loss=None, optimizer='sgd')
model.summary()
# train
import numpy as np
a = np.random.randn(8,32,32,3)
a_true = np.random.randn(8,10)
a_is_weight = np.random.randint(0,2,size=(8,1))
def __init__(
self,
n_filters=256,
# the number of dubs is inferred from the code not the paper
n_dubs=6,
n_convs_recon=9,
n_scales=3,
convs_per_scale=CONVS_PER_SCALE_DUB,
**kwargs):
super(DIDN, self).__init__(**kwargs)
self.n_filters = n_filters
self.n_dubs = n_dubs
self.n_scales = n_scales
self.convs_per_scale = convs_per_scale
self.n_convs_recon = n_convs_recon
self.dubs = [
DUB(
n_filters=self.n_filters,
n_scales=self.n_scales,
convs_per_scale=self.convs_per_scale,
) for _ in range(self.n_dubs)
]
self.recon_block = ReconBlock(
n_convs=self.n_convs_recon,
n_filters=self.n_filters,
)
self.first_conv = Conv2D(
filters=self.n_filters,
kernel_size=3,
padding='same',
activation=PReLU(shared_axes=[1, 2]),
use_bias=False,
)
self.pooling = Conv2D(
filters=self.n_filters,
kernel_size=3,
strides=2,
padding='same',
use_bias=False,
)
self.post_recon_agg = Conv2D(
filters=self.n_filters,
kernel_size=1,
padding='same',
use_bias=False,
)
self.post_recon_conv = Conv2D(
filters=self.n_filters,
kernel_size=3,
padding='same',
activation=PReLU(shared_axes=[1, 2]),
use_bias=False,
)
self.last_conv = Conv2D(
filters=1,
kernel_size=3,
padding='same',
# in code the activation is actually linear
activation='linear',
use_bias=False,
)
def _conv_layer(filters, kernel_size, strides=(1, 1), padding='same', name=None):
return Conv2D(filters, kernel_size, strides=strides, padding=padding,
use_bias=True, kernel_initializer='he_normal', name=name)
