# -*- coding: utf-8 -*-
import os
from tensorflow_core.python.keras.datasets import cifar10
from tensorflow_core.python.keras.engine.input_layer import Input
from tensorflow_core.python.keras.layers.convolutional import Conv2D
from tensorflow_core.python.keras.layers.core import Dense, Dropout, Activation, Flatten
from tensorflow_core.python.keras.layers.normalization import BatchNormalization
from tensorflow_core.python.keras.layers.pooling import MaxPooling2D
from tensorflow_core.python.keras.models import Sequential, Model, load_model
from tensorflow_core.python.keras.optimizer_v2.adam import Adam
import tensorflow as tf
from train_and_evaluation import evaluate_regression_model, train_model
from datasets import load_house_dataset_data, DatasetType
import matplotlib.pyplot as plt
__author__ = 106360
def generate_simple_cnn_regression_model(input_shape,n_blocks=3,weights='',is_regression=True,num_classes=1,freeze=False,remove_head=False):
# define the model input
inputs = Input(shape=(input_shape,input_shape,3))
# loop over the number of filters
x = inputs
for n in range(n_blocks):
x = Conv2D(32, (3, 3), padding="same",name='conv_%d' % n)(x)
x = Activation("relu")(x)
# x = BatchNormalization()(x)
x = MaxPooling2D()(x)
x= Flatten()(x)
x = Dense(16, activation="relu", name='pre_last_dense_reg')(x)
if not remove_head:
if is_regression:
y = Dense(num_classes, activation="sigmoid",name='last_dense_reg')(x)
else:
y = Dense(num_classes, activation="softmax", name='last_dense_clf')(x)
else:
y=x
model = Model(inputs,y)
if weights!='':
model.load_weights(weights,by_name=True)
if freeze:
for layer in model.layers[:-2]:
layer.trainable = False
return model
def train_cifar100(num_classes=100,batch_size=32):
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
y_train = tf.keras.utils.to_categorical(y_train, num_classes)
y_test = tf.keras.utils.to_categorical(y_test, num_classes)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('Using real-time data augmentation.')
# This will do preprocessing and realtime data augmentation:
datagen = tf.keras.preprocessing.image.ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset
samplewise_center=False, # set each sample mean to 0
featurewise_std_normalization=False, # divide inputs by std of the dataset
samplewise_std_normalization=False, # divide each input by its std
zca_whitening=False, # apply ZCA whitening
zca_epsilon=1e-06, # epsilon for ZCA whitening
rotation_range=0, # randomly rotate images in the range (degrees, 0 to 180)
# randomly shift images horizontally (fraction of total width)
width_shift_range=0.1,
# randomly shift images vertically (fraction of total height)
height_shift_range=0.1,
shear_range=0., # set range for random shear
zoom_range=0., # set range for random zoom
channel_shift_range=0., # set range for random channel shifts
# set mode for filling points outside the input boundaries
fill_mode='nearest',
cval=0., # value used for fill_mode = "constant"
horizontal_flip=True, # randomly flip images
vertical_flip=False, # randomly flip images
# set rescaling factor (applied before any other transformation)
rescale=None,
# set function that will be applied on each input
preprocessing_function=None,
# image data format, either "channels_first" or "channels_last"
data_format=None,
# fraction of images reserved for validation (strictly between 0 and 1)
validation_split=0.0)
# Compute quantities required for feature-wise normalization
# (std, mean, and principal components if ZCA whitening is applied).
datagen.fit(x_train)
dataset_train = datagen.flow(x_train, y_train, batch_size=batch_size)
dataset_test = datagen.flow(x_test, y_test, batch_size=batch_size)
num_steps_train = x_train.shape[0] // batch_size
num_steps_test = y_train.shape[0] // batch_size
model = generate_simple_cnn_regression_model(32, is_regression=False, num_classes=100)
opt = Adam(lr=1e-3, decay=1e-3 / 200)
model.compile(loss='categorical_crossentropy',
metrics=['categorical_crossentropy', 'accuracy'],
optimizer=opt)
model.summary()
history = model.fit_generator(dataset_train, steps_per_epoch=num_steps_train,
epochs=20,
validation_data=(x_test, y_test),
workers=1)
return model
if __name__ == "__main__":
pre_train_with_cifar100 = True
weights = ''
if pre_train_with_cifar100:
file_weight_cifar100 = 'pretrained_cifar100.h5'
try:
model = load_model(file_weight_cifar100)
except:
model = train_cifar100()
model.save(file_weight_cifar100)
weights = file_weight_cifar100
(trainX_data,trainX_img, trainY, testX_data,testX_img,testY), normalizer = load_house_dataset_data(test_size=0.2,random_state=666,type=DatasetType.Both)
trainX = trainX_img['bathroom_img']
testX = testX_img['bathroom_img']
input_shape = trainX.shape[1]
if pre_train_with_cifar100:
file_weight_finetune = 'regression_model_image_finetune.h5'
model = generate_simple_cnn_regression_model(input_shape,weights=weights,freeze=True)
opt = Adam(lr=1e-3, decay=1e-3 / 200)
model.compile(loss='mean_squared_error',metrics=['mean_absolute_percentage_error','mean_absolute_error','mean_squared_error'], optimizer=opt)
model.summary()
model = train_model(trainX, trainY, testX, testY,model,show_plot=True,epochs=500,batch_size=32)
evaluate_regression_model(model,testX,testY,normalizer,show_plot=True)
model.save(file_weight_finetune)
weights = file_weight_finetune
final_model_weight = 'regression_model_image_pretrained.h5'
else:
final_model_weight = 'regression_model_image_from_scratch.h5'
model = generate_simple_cnn_regression_model(input_shape, weights=weights)
opt = Adam(lr=1e-3, decay=1e-3 / 200)
model.compile(loss='mean_squared_error',
metrics=['mean_absolute_percentage_error', 'mean_absolute_error', 'mean_squared_error'],
optimizer=opt)
model.summary()
model = train_model(trainX, trainY, testX, testY, model, show_plot=True, epochs=500, batch_size=32)
evaluate_regression_model(model, testX, testY, normalizer, show_plot=True)
model.save(final_model_weight)