def build_model():
model = Sequential()
model.add(RandomRotation(factor=0.45))
model.add(InputLayer(input_shape=(224, 224, 1)))
model.add(Conv2D(1, (2, 2), activation='relu', padding='same'))
model.add(Conv2D(32, (2, 2), strides=2, activation='relu', padding='same'))
model.add(Conv2D(64, (2, 2), strides=2, activation='relu', padding='same'))
model.add(Conv2D(128, (2, 2), strides=2, activation='relu',
padding='same'))
model.add(
Conv2DTranspose(128, (2, 2),
strides=2,
activation='relu',
padding='same'))
model.add(
Conv2DTranspose(64, (2, 2),
strides=2,
activation='relu',
padding='same'))
model.add(
Conv2DTranspose(2, (2, 2),
strides=2,
activation='relu',
padding='same'))
return model
def create_and_train_model():
#Importing data
mnist = keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
#Normalizing the input
x_train = keras.utils.normalize(x_train, axis=1)
x_test = keras.utils.normalize(x_test, axis=1)
#reshape the arrays for the CNN
x_train_r = np.expand_dims(x_train, axis=3)
x_test_r = np.expand_dims(x_test, axis=3)
#Translating and Rotating the training set
data_augmentation = Sequential([
RandomTranslation(0.05, 0.05, fill_mode='constant'),
RandomRotation(0.02, fill_mode='constant')
])
x_train_r = data_augmentation(x_train_r)
model = Sequential()
model.add(
Conv2D(32,
kernel_size=5,
padding='same',
activation='relu',
input_shape=(28, 28, 1)))
model.add(MaxPooling2D())
model.add(Dropout(0.2))
model.add(Conv2D(64, kernel_size=5, padding='same', activation='relu'))
model.add(MaxPooling2D(padding='same'))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(10, activation='softmax'))
model.compile(optimizer="adam",
loss="sparse_categorical_crossentropy",
metrics=["accuracy"])
monitor = EarlyStopping(monitor='val_loss',
min_delta=1e-3,
patience=5,
verbose=1,
mode='auto',
restore_best_weights=True)
model.fit(x_train_r,
y_train,
validation_data=(x_test_r, y_test),
callbacks=[monitor],
verbose=1,
epochs=20)
model.save('number_recognition_CNN')
def preprocessing_layers(a, hp):
#a = preprocess_input(a)#Rescaling(1/255)(a)
translate = hp.Float("translate", 0, 0.5, step=0.1, default=0.2)
a = RandomTranslation(
#hp.Float("t_x", 0, 0.5, step=0.1, default=0.2),
#hp.Float("t_y", 0, 0.5, step=0.1, default=0.2),
translate,
translate,
fill_mode="reflect",interpolation="bilinear",)(a)
a = RandomFlip()(a)
a = RandomZoom(hp.Float("zoom", 0, 0.5, step=0.1, default=0.25))(a)
a = RandomRotation(hp.Float("rotation", 0, 2, step=0.1, default=2))(a)
#a = RandomHeight(0.2)(a)
#a = RandomWidth(0.2)(a)
return a
def get_data(path, num_classes):
#the value passed in split to get training and testing is as per the requirement mentioned
#for caltech101 and caltech256 dataset it gets randomly 30 images from class for training and remaining for testing
#fetch training data
train = tf.keras.preprocessing.image_dataset_from_directory(
directory=path,
labels="inferred",
label_mode="int",
validation_split=0.6654,
subset="training",
seed=123,
image_size=(224, 224),
batch_size=1)
#fetch testing data
test = tf.keras.preprocessing.image_dataset_from_directory(
directory=path,
labels="inferred",
label_mode="int",
validation_split=0.6654,
subset="validation",
seed=123,
image_size=(224, 224),
batch_size=1)
#noramlize the data
normalization = Rescaling(1. / 255)
#data agumentation
train_aug = Sequential([
Rescaling(1. / 255),
RandomFlip("horizontal_and_vertical"),
RandomRotation(0.2)
])
#normalize train and test data apply one hot encoding on labels and squeeze train and test to its required dimension
#apply train_aug on train set and normalization only on test set
#becomes equal to the shape of the input to the convolution network
train_n = train.map(lambda x, y: (tf.squeeze(train_aug(x)),
tf.squeeze(tf.one_hot(y, num_classes))))
test_n = test.map(lambda x, y: (tf.squeeze(normalization(x)),
tf.squeeze(tf.one_hot(y, num_classes))))
#the output of train_n and test_n is mapped dataset so we iterate over the train_n and test_n
#and store the features and labels in list
train_inputs = [] #for training features
train_labels = [] #for training labels
test_inputs = [] #for testing features
test_labels = [] #for testing labels
c = 0
tr = iter(train_n) #create train_n iterator
tt = iter(test_n) #create test_n iterator
for i in range(0, len(test_n)):
if c < len(train_n):
i, j = next(tr)
train_inputs.append(i)
train_labels.append(j)
c += 1
i, j = next(tt)
test_inputs.append(i)
test_labels.append(j)
#convert the list to tensor
train_inputs = tf.convert_to_tensor(train_inputs, dtype=tf.float32)
train_labels = tf.convert_to_tensor(train_labels, dtype=tf.float32)
test_inputs = tf.convert_to_tensor(test_inputs, dtype=tf.float32)
test_labels = tf.convert_to_tensor(test_labels, dtype=tf.float32)
#free memory
del train_n, test_n, tr, tt, train, test
return train_inputs, train_labels, test_inputs, test_labels
#hp.Float("t_x", 0, 0.5, step=0.1, default=0.2),
#hp.Float("t_y", 0, 0.5, step=0.1, default=0.2),
translate,
translate,
fill_mode="reflect",interpolation="bilinear",)(a)
a = RandomFlip()(a)
a = RandomZoom(hp.Float("zoom", 0, 0.5, step=0.1, default=0.25))(a)
a = RandomRotation(hp.Float("rotation", 0, 2, step=0.1, default=2))(a)
#a = RandomHeight(0.2)(a)
#a = RandomWidth(0.2)(a)
return a
data_augmentation = Sequential([
RandomFlip("horizontal_and_vertical"),
RandomRotation((0,0.5), fill_mode='constant')
])
def comparisons_model(hp):
img_size=224
weights=None
"""
Create comparisons network which reproduce the choice in an images duel.
:param img_size: size of input images during training
:type img_size: tuple(int)
:param weights: path to the weights use for initialization
:type weights: str
:return: ranking comparisons model
:rtype: keras.Model
"""
# X_train,x_test,Y_train,Y_test = train_test_split(X_train,Y_train,test_size=0.2, random_state = 42)
# One hot coding
from sklearn import preprocessing
lb = preprocessing.LabelBinarizer()
Y_train = lb.fit_transform(Y_train)
class_num = Y_train.shape[1]
# print(Y_train)
# Creating the model
model = Sequential()
# Adding data augmentation
model.add(RandomContrast([0.5,1.5]))
model.add(RandomFlip('horizontal_and_vertical'))
model.add(RandomRotation([0.5,1.5]))
model.add(Resizing(128,128,interpolation='bicubic'))
model.add(RandomZoom([0.2,0.5]))
model.add(RandomTranslation(height_factor=(-0.2,0.2), width_factor=(-0.2,0.2)))
#first convolutional layer
model.add(Conv2D(32, (5,5), input_shape=(48,48,1), activation='relu', padding='same'))
# model.add(Conv2D(32, (5,5), input_shape=(48,48,1), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(BatchNormalization())
#second convolutional layer
model.add(Conv2D(64,(5,5), activation='relu', padding='same'))
model.add(Conv2D(64,(5,5), activation='relu', padding='same'))
# model.add(Conv2D(64,(5,5), activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2,2)))