cifar100_InceptionV3.py

# datasetを読み込む
from keras.datasets import cifar100 
(x_train, y_train), (x_test, y_test) = cifar100.load_data()

# 入力のリサイズ (InceptionV3の仕様のため)
import cv2
import numpy as np
from keras.utils import np_utils

#image_size = 299
image_size = 75

def resize(data):
    return np.array([cv2.resize(img, dsize=(image_size, image_size)) for img in data])

x_train = resize(x_train)
y_train = np_utils.to_categorical(y_train)

x_test= resize(x_test)
y_test = np_utils.to_categorical(y_test)

# 学習するモデルを設定
from keras.applications import InceptionV3
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D

num_classes = 100

base_model = InceptionV3(include_top=False, weights='imagenet', input_shape=(image_size, image_size, 3))

x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)

model = Model(inputs=base_model.input, outputs=predictions)

for layer in model.layers[:249]:
    layer.trainable = False

    if layer.name.startswith('batch_normalization'):
        layer.trainable = True
for layer in model.layers[249:]:
    layer.trainable = True

# モデルのコンパイル
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# 学習に関する設定と学習
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import EarlyStopping, ModelCheckpoint

data_generator = ImageDataGenerator(
    featurewise_center = False,
    samplewise_center = False,
    featurewise_std_normalization = False,
    samplewise_std_normalization = False,
    zca_whitening = False,
    rotation_range = 0,
    width_shift_range = 0.1,
    height_shift_range = 0.1,
    horizontal_flip = True,
    vertical_flip = False
)

early_stopping = EarlyStopping(monitor='val_loss', patience=5)
checkpoint = ModelCheckpoint(filepath = 'chpt.hdf5', monitor='val_loss', verbose=1, save_best_only=True)

history = model.fit_generator(
    data_generator.flow(x_train, y_train, batch_size = 32),
    steps_per_epoch = x_train.shape[0] // 32,
    validation_data = (x_test, y_test),
    verbose = 1,
    callbacks=[early_stopping, checkpoint],
    epochs=100
)

epochs = early_stopping.stopped_epoch

# 学習過程のグラフを出力
import matplotlib.pyplot as plt

def plot_history(history):
    # accuracy
    plt.plot(history.history['acc'])
    plt.plot(history.history['val_acc'])
    plt.title('model accuracy')
    plt.xlabel('epoch')
    plt.ylabel('accuracy')
    plt.legend(['acc', 'val_acc'], loc='lower right')
    plt.savefig('acc.png')
    plt.clf()

    # loss
    plt.plot(history.history['loss'])
    plt.plot(history.history['val_loss'])
    plt.title('model loss')
    plt.xlabel('epoch')
    plt.ylabel('loss')
    plt.legend(['loss', 'val_loss'], loc='lower right')
    plt.savefig('loss.png')

plot_history(history)

# モデルの保存
model.save('model.h5')

# historyの保存
import json
with open('history.json', 'w') as f:
  json.dump(history.history, f)

score = model.evaluate(x_test, y_test, verbose=1)
print('Loss:', score[0])
print('Accuracy:', score[1])