版本:Tensordlow2.1
运行环境:Google Colaboratory
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import keras_applications
import tensorflow as tf
from tensorflow.keras import datasets, Sequential, layers
from tensorflow.keras.layers import Conv2D, MaxPool2D, Dense, Flatten, Dropout
import matplotlib.pyplot as plt
# datasets download
(x, y), (x_val, y_val) = datasets.cifar10.load_data()
print(x.shape, y.shape)
def preprocess(x, y):
"""
:param x:
:param y:
:return:
"""
x = tf.cast(x, dtype=tf.float32) / 255.
x = tf.reshape(x, [-1, 32, 32, 3])
y = tf.one_hot(tf.cast(y, dtype=tf.int32), depth=10)
y = tf.reshape(y, [-1, 10])
return x, y
train_db = tf.data.Dataset.from_tensor_slices((x, y))
val_db = tf.data.Dataset.from_tensor_slices((x_val, y_val))
train_db = train_db.batch(256).map(preprocess)
val_db = val_db.map(preprocess)
sample = next(iter(train_db))
print(sample[0].shape, sample[1].shape)
model = Sequential([
# unit1
Conv2D(filters=64, kernel_size=(3, 3), activation='relu', padding='same', name='Block1_conv1',
input_shape=(32, 32, 3)),
Conv2D(filters=64, kernel_size=(3, 3), activation='relu', padding='same', name='Block1_conv2'),
MaxPool2D(pool_size=(2, 2), strides=(2, 2), name='Block1_pool', padding='same'),
# unit2
Conv2D(filters=128, kernel_size=(3, 3), activation='relu', padding='same', name='Block2_conv1'),
Conv2D(filters=123, kernel_size=(3, 3), activation='relu', padding='same', name='Block2_conv2'),
MaxPool2D(pool_size=(2, 2), strides=(2, 2), name='Block2_pool', padding='same'),
# unit3
Conv2D(filters=256, kernel_size=(3, 3), activation='relu', padding='same', name='Block3_conv1'),
Conv2D(filters=256, kernel_size=(3, 3), activation='relu', padding='same', name='Block3_conv2'),
MaxPool2D(pool_size=(2, 2), strides=(2, 2), name='Block3_pool', padding='same'),
# unit4
Conv2D(filters=512, kernel_size=(3, 3), activation='relu', padding='same', name='Block4_conv1'),
Conv2D(filters=512, kernel_size=(3, 3), activation='relu', padding='same', name='Block4_conv2'),
MaxPool2D(pool_size=(2, 2), strides=(2, 2), name='Block4_pool', padding='same'),
# unit5
Conv2D(filters=512, kernel_size=(3, 3), activation='relu', padding='same', name='Block5_conv1'),
Conv2D(filters=512, kernel_size=(3, 3), activation='relu', padding='same', name='Block5_conv2'),
MaxPool2D(pool_size=(2, 2), strides=(2, 2), name='Block5_pool', padding='same'),
])
model.add(Flatten()) # 扁平化
model.add(Dense(4096, activation='relu'))
model.add(Dropout(rate=0.5))
model.add(Dense(4096, activation='relu'))
model.add(Dropout(rate=0.5))
model.add(Dense(10, activation='softmax'))
model.summary()
model.compile(optimizer='sgd', loss='categorical_crossentropy', metrics=['accuracy'])
history = model.fit(train_db, epochs=20, validation_data=val_db, shuffle=True)
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('AlexNet')
plt.xlabel('epoch')
plt.ylabel('loss')
plt.legend(['Train', 'loss'])
plt.show()
输出:
Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
170500096/170498071 [==============================] - 2s 0us/step
(50000, 32, 32, 3) (50000, 1)
(256, 32, 32, 3) (256, 10)
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
Block1_conv1 (Conv2D) (None, 32, 32, 64) 1792
_________________________________________________________________
Block1_conv2 (Conv2D) (None, 32, 32, 64) 36928
_________________________________________________________________
Block1_pool (MaxPooling2D) (None, 16, 16, 64) 0
_________________________________________________________________
Block2_conv1 (Conv2D) (None, 16, 16, 128) 73856
_________________________________________________________________
Block2_conv2 (Conv2D) (None, 16, 16, 123) 141819
_________________________________________________________________
Block2_pool (MaxPooling2D) (None, 8, 8, 123) 0
_________________________________________________________________
Block3_conv1 (Conv2D) (None, 8, 8, 256) 283648
_________________________________________________________________
Block3_conv2 (Conv2D) (None, 8, 8, 256) 590080
_________________________________________________________________
Block3_pool (MaxPooling2D) (None, 4, 4, 256) 0
_________________________________________________________________
Block4_conv1 (Conv2D) (None, 4, 4, 512) 1180160
_________________________________________________________________
Block4_conv2 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
Block4_pool (MaxPooling2D) (None, 2, 2, 512) 0
_________________________________________________________________
Block5_conv1 (Conv2D) (None, 2, 2, 512) 2359808
_________________________________________________________________
Block5_conv2 (Conv2D) (None, 2, 2, 512) 2359808
_________________________________________________________________
Block5_pool (MaxPooling2D) (None, 1, 1, 512) 0
_________________________________________________________________
flatten (Flatten) (None, 512) 0
_________________________________________________________________
dense (Dense) (None, 4096) 2101248
_________________________________________________________________
dropout (Dropout) (None, 4096) 0
_________________________________________________________________
dense_1 (Dense) (None, 4096) 16781312
_________________________________________________________________
dropout_1 (Dropout) (None, 4096) 0
_________________________________________________________________
dense_2 (Dense) (None, 10) 40970
=================================================================
Total params: 28,311,237
Trainable params: 28,311,237
Non-trainable params: 0
_________________________________________________________________
Epoch 1/20
196/196 [==============================] - 101s 517ms/step - loss: 2.3025 - accuracy: 0.1031 - val_loss: 2.3024 - val_accuracy: 0.1233
Epoch 2/20
196/196 [==============================] - 100s 509ms/step - loss: 2.3023 - accuracy: 0.1174 - val_loss: 2.3021 - val_accuracy: 0.1120
Epoch 3/20
196/196 [==============================] - 100s 512ms/step - loss: 2.3020 - accuracy: 0.1268 - val_loss: 2.3018 - val_accuracy: 0.1225
Epoch 4/20
196/196 [==============================] - 100s 510ms/step - loss: 2.3017 - accuracy: 0.1415 - val_loss: 2.3014 - val_accuracy: 0.1603
Epoch 5/20
196/196 [==============================] - 100s 510ms/step - loss: 2.3013 - accuracy: 0.1579 - val_loss: 2.3009 - val_accuracy: 0.1721
Epoch 6/20
196/196 [==============================] - 101s 514ms/step - loss: 2.3007 - accuracy: 0.1676 - val_loss: 2.3002 - val_accuracy: 0.1944
Epoch 7/20
196/196 [==============================] - 100s 510ms/step - loss: 2.2999 - accuracy: 0.1765 - val_loss: 2.2991 - val_accuracy: 0.1984
Epoch 8/20
196/196 [==============================] - 100s 510ms/step - loss: 2.2987 - accuracy: 0.1815 - val_loss: 2.2974 - val_accuracy: 0.2104
Epoch 9/20
196/196 [==============================] - 101s 513ms/step - loss: 2.2968 - accuracy: 0.1867 - val_loss: 2.2946 - val_accuracy: 0.2222
Epoch 10/20
196/196 [==============================] - 100s 510ms/step - loss: 2.2936 - accuracy: 0.1895 - val_loss: 2.2897 - val_accuracy: 0.2388
Epoch 11/20
196/196 [==============================] - 100s 510ms/step - loss: 2.2878 - accuracy: 0.1881 - val_loss: 2.2800 - val_accuracy: 0.2178
Epoch 12/20
196/196 [==============================] - 101s 514ms/step - loss: 2.2741 - accuracy: 0.1829 - val_loss: 2.2539 - val_accuracy: 0.1891
Epoch 13/20
196/196 [==============================] - 100s 510ms/step - loss: 2.2216 - accuracy: 0.1727 - val_loss: 2.1703 - val_accuracy: 0.1859
Epoch 14/20
196/196 [==============================] - 100s 510ms/step - loss: 2.1630 - accuracy: 0.1827 - val_loss: 2.1334 - val_accuracy: 0.2134
Epoch 15/20
196/196 [==============================] - 101s 515ms/step - loss: 2.1330 - accuracy: 0.1907 - val_loss: 2.0949 - val_accuracy: 0.2240
Epoch 16/20
196/196 [==============================] - 100s 510ms/step - loss: 2.1070 - accuracy: 0.2051 - val_loss: 2.0692 - val_accuracy: 0.2306
Epoch 17/20
196/196 [==============================] - 100s 511ms/step - loss: 2.0787 - accuracy: 0.2187 - val_loss: 2.0274 - val_accuracy: 0.2538
Epoch 18/20
196/196 [==============================] - 101s 514ms/step - loss: 2.0499 - accuracy: 0.2319 - val_loss: 1.9917 - val_accuracy: 0.2707
Epoch 19/20
196/196 [==============================] - 100s 510ms/step - loss: 2.0226 - accuracy: 0.2437 - val_loss: 1.9686 - val_accuracy: 0.2793
Epoch 20/20
196/196 [==============================] - 100s 510ms/step - loss: 1.9990 - accuracy: 0.2541 - val_loss: 1.9402 - val_accuracy: 0.2896
可能是由于训练的时间较短,所以效果并不明显;