直接上程式碼

from keras.preprocessing.image import ImageDataGenerator
from keras import optimizers
import keras
import os
from keras import models
from keras import layers
from tensorflow.keras.callbacks import TensorBoard
from keras.applications import VGG16

original_dataset_dir = '/home/ysy/tf_keras_learning/kagglecatsanddogs_3367a (1)'

# Directories for our training,
# validation and test splits
train_dir = os.path.join(original_dataset_dir, 'train')
#os.mkdir(train_dir)
validation_dir = os.path.join(original_dataset_dir, 'validation')
#os.mkdir(validation_dir)
test_dir = os.path.join(original_dataset_dir, 'test')
#os.mkdir(test_dir)

# Directory with our training cat pictures
train_cats_dir = os.path.join(train_dir, 'cats')
#os.mkdir(train_cats_dir)

# Directory with our training dog pictures
train_dogs_dir = os.path.join(train_dir, 'dogs')
#os.mkdir(train_dogs_dir)

# Directory with our validation cat pictures
validation_cats_dir = os.path.join(validation_dir, 'cats')
#os.mkdir(validation_cats_dir)

# Directory with our validation dog pictures
validation_dogs_dir = os.path.join(validation_dir, 'dogs')
#os.mkdir(validation_dogs_dir)

# Directory with our validation cat pictures
test_cats_dir = os.path.join(test_dir, 'cats')
#os.mkdir(test_cats_dir)

# Directory with our validation dog pictures
test_dogs_dir = os.path.join(test_dir, 'dogs')
#os.mkdir(test_dogs_dir)


conv_base = VGG16(weights='imagenet',
                  include_top=False,
                  input_shape=(150, 150, 3))

NAME = "VGG-16_pretrain_1"
print(NAME)


model = models.Sequential()
model.add(conv_base)
model.add(layers.Flatten())
model.add(layers.Dense(256, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

conv_base.trainable = True

set_trainable = False

print('trainable weights is :',len(model.trainable_weights))

train_datagen = ImageDataGenerator(
      rescale=1./255,
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')

test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
      train_dir,
      target_size=(150, 150),
      batch_size=20,
      class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

tensorboard = TensorBoard(log_dir="logs/{}".format(NAME))

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=2e-5),
              metrics=['acc'])

history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=50,
      verbose=2)

model.save('VGG16_pretrain_all.model')
 

訓練結果

Epoch 28/30
 - 29s - loss: 0.0979 - acc: 0.9635 - val_loss: 0.1180 - val_acc: 0.9630
Epoch 29/30
 - 29s - loss: 0.1071 - acc: 0.9585 - val_loss: 0.0908 - val_acc: 0.9675
Epoch 30/30
 - 29s - loss: 0.1029 - acc: 0.9555 - val_loss: 0.1363 - val_acc: 0.9605

測試程式碼

import sys
import os
from keras.preprocessing.image import ImageDataGenerator

sys.path.remove("/opt/ros/kinetic/lib/python2.7/dist-packages")

import cv2
import tensorflow as tf

original_dataset_dir = '/home/ysy/tf_keras_learning/kagglecatsanddogs_3367a (1)'
test_dir = os.path.join(original_dataset_dir, 'test')

CATEGORIES = ["Dog", "Cat"]


def prepare(filepath):
    IMG_SIZE = 150
    new_array = cv2.resize(img_array, (IMG_SIZE, IMG_SIZE))
    return new_array.reshape(-1, IMG_SIZE, IMG_SIZE, 1)

model = tf.keras.models.load_model("VGG16_pretrain_all.model")

test_datagen = ImageDataGenerator(rescale=1./255)
test_generator = test_datagen.flow_from_directory(
        test_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

test_loss, test_acc = model.evaluate_generator(test_generator, steps=50)

#prediction = model.predict([prepare('/home/ysy/tf_keras_learning/cat2.png')])

print('test acc:', test_acc)

#print(prediction)
#print(CATEGORIES[int(prediction[0][0])])
 

測試結果

Found 2000 images belonging to 2 classes.
test acc: 0.9649999928474426