https://employment.en-japan.com/engineerhub/entry/2017/04/28/110000#3-Inception-v3%E3%83%A2%E3%83%87%E3%83%AB%E3%82%92%E7%94%A8%E3%81%84%E3%81%9F%E7%94%BB%E5%83%8F%E8%AA%8D%E8%AD%98 この辺を参考に

fine_tuning_inceptionv3.py

from keras.applications.inception_v3 import InceptionV3
from keras.applications.inception_v3 import preprocess_input
from keras.models import Sequential, Model
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution2D, MaxPooling2D, ZeroPadding2D, GlobalAveragePooling2D, AveragePooling2D
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import ModelCheckpoint, CSVLogger, LearningRateScheduler, ReduceLROnPlateau
from keras.optimizers import SGD
from keras.regularizers import l2
import matplotlib.image as mpimg
from scipy.misc import imresize
import numpy as np
import keras.backend as K
import math

K.clear_session()
img_size=299
train_data_dir = "data/train"
validation_data_dir = "data/validation"

#訓練データ拡張
train_datagen = ImageDataGenerator(
        featurewise_center=False,
        samplewise_center=False,
        featurewise_std_normalization=False,
        samplewise_std_normalization=False,
        rotation_range=10,
        width_shift_range=0.2,
        height_shift_range=0.2,
        horizontal_flip=True,
        vertical_flip=False,
        zoom_range=[.8, 1],
        channel_shift_range=30,
        fill_mode='reflect')

test_datagen = ImageDataGenerator()

train_generator = train_datagen.flow_from_directory(
        train_data_dir,
        target_size=(img_size, img_size),
        batch_size=64,
        class_mode="binary")
test_generator = test_datagen.flow_from_directory(
        validation_data_dir,
        target_size=(img_size, img_size),
        batch_size=32,
        class_mode = "binary")

#Inception v3モデルの読み込み。最終層は読み込まない
base_model = InceptionV3(weights='imagenet', include_top=False)
#最終層の設定
x = base_model.output
x = GlobalAveragePooling2D()(x)
predictions = Dense(1, kernel_initializer="glorot_uniform", activation="sigmoid", kernel_regularizer=l2(.0005))(x)

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

#base_modelはweightsを更新しない
for layer in base_model.layers:
    layer.trainable = False

opt = SGD(lr=.01, momentum=.9)
model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])

checkpointer = ModelCheckpoint(filepath='model.{epoch:02d}-{val_loss:.2f}.hdf5', verbose=1, save_best_only=True)
csv_logger = CSVLogger('model.log')

reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.2,
                  patience=5, min_lr=0.001)

history = model.fit_generator(train_generator,
                    steps_per_epoch=2000,
                    epochs=10,
                    validation_data=test_generator,
                    validation_steps=800,
                    verbose=1,
                    callbacks=[reduce_lr, csv_logger, checkpointer])

predict.py

from keras.models import load_model
from keras.preprocessing import image
import numpy as np

img_size=299
model = load_model(filepath='./model.00-0.06.hdf5')
def predict_img(img_name):
    img_arr = image.img_to_array(image.load_img(img_name, target_size=(img_size, img_size)))
    x=img_arr
    x = np.expand_dims(x, axis=0)
    y_pred = model.predict(x)
    if y_pred <0.5:
        print (y_pred[0][0], 'cat')
    else:
        print (y_pred[0][0], 'dog')