def make_model(shape):
input = Input(shape)
cur_layer = Conv2D(3, (3, 3), activation='relu', padding='same')(input)
cur_layer = Conv2D(3, (5, 5), activation='relu', padding='same')(cur_layer)
cur_layer = MaxPool2D(pool_size=(3, 3), strides=(3, 3))(cur_layer)
cur_layer = BatchNormalization()(cur_layer)
cur_layer = Dropout(0.5)(cur_layer)
cur_layer = Conv2D(16, (5, 5), activation='relu',
padding='same')(cur_layer)
cur_layer = MaxPool2D(pool_size=(5, 5), strides=(5, 5))(cur_layer)
cur_layer = BatchNormalization()(cur_layer)
cur_layer = Dropout(0.5)(cur_layer)
cur_layer = Flatten()(cur_layer)
o1 = Dense(310, name='f')(cur_layer)
o2 = concatenate([o1, cur_layer])
o2 = Dense(3678, name='g')(o2)
o3 = concatenate([o1, o2, cur_layer])
o3 = Dense(32094, name='c')(o3)
cur_layer = Model(inputs=input, outputs=[o1, o2, o3])
opt = Adam(lr=0.001, amsgrad=True)
cur_layer.compile(optimizer=opt,
loss=[
'sparse_categorical_crossentropy',
'sparse_categorical_crossentropy',
'sparse_categorical_crossentropy'
],
metrics=['accuracy'])
return cur_layer
def generate_classifier(size=128, classes=4, learning_rate=0.01):
x, y = size, size #pixel_crop*number_of_crops,pixel_crop*number_of_crops
i = Input(shape=(x, y, 3))
m = Conv2D(64, kernel_size=(3, 3), padding="same", activation="relu")(i)
m = BatchNormalization()(m)
m = MaxPool2D()(m)
m = Conv2D(64, kernel_size=(3, 3), padding="same", activation="relu")(m)
m = BatchNormalization()(m)
m = MaxPool2D()(m)
m = Conv2D(128, kernel_size=(3, 3), padding="same", activation="relu")(m)
m = BatchNormalization()(m)
m = MaxPool2D()(m)
m = Conv2D(256, kernel_size=(3, 3), padding="same", activation="relu")(m)
m = BatchNormalization()(m)
m = MaxPool2D()(m)
m = Conv2D(512, kernel_size=(3, 3), padding="same", activation="relu")(m)
m = BatchNormalization()(m)
m = MaxPool2D()(m)
m = Flatten()(m)
m = Dense(512, activation="relu")(m)
m = Dropout(0.3)(m)
out1 = Dense(classes, activation="softmax", name="output_classes")(m)
out2 = Dense(1, activation="sigmoid", name="output_score")(m)
m = Model(i, [out1, out2])
optimizer = Adam(learning_rate=learning_rate)
m.compile(optimizer=optimizer,
loss={
'output_score': 'mse',
'output_classes': 'categorical_crossentropy'
},
loss_weights={
'output_score': 5.,
'output_classes': 1.
})
return m
def get_model(img_size = 64, loss_weights=[0.5, 0.25, 0.25], optimizer='adam'):
inputs = Input(shape = (img_size, img_size, 1))
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(inputs)
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=32, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=64, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=128, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=256, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Flatten()(model)
model = Dense(1024, activation = "relu")(model)
model = Dropout(rate=0.3)(model)
dense = Dense(512, activation = "relu")(model)
head_root = Dense(168, activation = 'softmax', name='out_root')(dense)
head_vowel = Dense(11, activation = 'softmax', name='out_vowel')(dense)
head_consonant = Dense(7, activation = 'softmax', name='out_consonant')(dense)
model = Model(inputs=inputs, outputs=[head_root, head_vowel, head_consonant])
model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['acc'],
loss_weights=loss_weights)
return model
def create_model(img_size=64):
inputs = Input(shape=(img_size, img_size, 1))
model = Conv2D(filters=32,
kernel_size=(3, 3),
padding='SAME',
activation='relu',
input_shape=(img_size, img_size, 1))(inputs)
model = Conv2D(filters=32,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=32,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=32,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=32,
kernel_size=(5, 5),
padding='SAME',
activation='relu')(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=64,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=64,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=64,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=64,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=64,
kernel_size=(5, 5),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=128,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=128,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=128,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=128,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=128,
kernel_size=(5, 5),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Conv2D(filters=256,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=256,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=256,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = Conv2D(filters=256,
kernel_size=(3, 3),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=256,
kernel_size=(5, 5),
padding='SAME',
activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.3)(model)
model = Flatten()(model)
model = Dense(1024, activation="relu")(model)
model = Dropout(rate=0.3)(model)
dense = Dense(512, activation="relu")(model)
head_root = Dense(168, activation='softmax')(dense)
head_vowel = Dense(11, activation='softmax')(dense)
head_consonant = Dense(7, activation='softmax')(dense)
model = Model(inputs=inputs,
outputs=[head_root, head_vowel, head_consonant])
model.summary()
plot_model(model, to_file='./results/model.png')
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
return model
plt.grid(False)
plt.imshow(train_images_cifar[i], cmap=plt.cm.binary)
# The CIFAR labels happen to be arrays,
# which is why you need the extra index
plt.xlabel(class_names[train_labels_cifar[i][0]])
plt.show()
# Learning rate tested with 0.001, 0.005 and 0.01 over 300 epochs for Table VII
sgd = optimizers.SGD(learning_rate=0.005,
decay=1e-6,
momentum=0.9,
nesterov=True)
model.compile(
optimizer=sgd,
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
# Batch size of 128, 256 & 512 tested for Modification 2
# Batch size set to 128 for Modification 3 and 4
history = model.fit(train_images_cifar,
train_labels_cifar,
batch_size=128,
epochs=200,
validation_data=(test_images_cifar, test_labels_cifar))
plt.plot(history.history["accuracy"])
plt.plot(history.history['val_accuracy'])
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
def create_model(image_size=64, weights_path='weights/bengalimodal.h5', pretrained=True):
"""
Description
-----------
Creates model from https://www.kaggle.com/amanmishra4yearbtech/bengali-classification-quick-implementation
Parameters
----------
image_size: int
Image size to use model with.
weights_path: str
Path to model's weights. Required if pretrained=True.
pretrained: bool:
Use pretrained weights or no.
Returns
-------
model: tensorflow.keras.models.Model
CNN model
"""
inputs = Input(shape=(image_size, image_size, 1))
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu',
input_shape=(image_size, image_size, 1))(inputs)
model = Conv2D(filters=32, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=32, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = Dropout(rate=0.25)(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=64, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=64, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.25)(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=128, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=128, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = Dropout(rate=0.2)(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = Conv2D(filters=256, kernel_size=(3, 3), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.15)(model)
model = MaxPool2D(pool_size=(2, 2))(model)
model = Conv2D(filters=256, kernel_size=(5, 5), padding='SAME', activation='relu')(model)
model = BatchNormalization(momentum=0.20)(model)
model = Dropout(rate=0.25)(model)
model = Flatten()(model)
model = Dense(512, activation="relu", name='dense_')(model)
model = Dropout(rate=0.25)(model)
dense = Dense(256, activation="relu", name='dense_1')(model)
head_root = Dense(168, activation='softmax', name='dense_2')(dense)
head_vowel = Dense(11, activation='softmax', name='dense_3')(dense)
head_consonant = Dense(7, activation='softmax', name='dense_4')(dense)
model = Model(inputs=inputs, outputs=[head_root, head_vowel, head_consonant]) # 3 outputs one for each
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
if pretrained:
model.load_weights(weights_path)
return model
