【tensorflow2.0】45.图片数据建模流程实例（总复习）

这次我们来看一下图片数据的建模流程，用到的数据集是cifar10的子数据及cifar2，分两个类别airplane和automobile，训练集有10000张图片，测试集有2000张。

import tensorflow as tf 
from tensorflow.keras import datasets,layers,models

BATCH_SIZE = 100
#定义图片预处理的函数
def load_image(img_path,size = (32,32)):
    label = tf.constant(1,tf.int8) if tf.strings.regex_full_match(img_path,".*automobile.*") \
            else tf.constant(0,tf.int8)
    img = tf.io.read_file(img_path)
    img = tf.image.decode_jpeg(img) #注意此处为jpeg格式
    #归一化
    img = tf.image.resize(img,size)/255.0
    return(img,label)

#使用并行化预处理num_parallel_calls 和预存数据prefetch来提升性能
ds_train = tf.data.Dataset.list_files("./data/cifar2/train/*/*.jpg") \
           .map(load_image, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
           .shuffle(buffer_size = 1000).batch(BATCH_SIZE) \
           .prefetch(tf.data.experimental.AUTOTUNE)  

ds_test = tf.data.Dataset.list_files("./data/cifar2/test/*/*.jpg") \
           .map(load_image, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
           .batch(BATCH_SIZE) \
           .prefetch(tf.data.experimental.AUTOTUNE)  

%matplotlib inline
%config InlineBackend.figure_format = 'svg'

#查看部分样本
from matplotlib import pyplot as plt 
plt.figure(figsize=(8,8)) 
for i,(img,label) in enumerate(ds_train.unbatch().take(9)):
    ax=plt.subplot(3,3,i+1)
    ax.imshow(img.numpy())
    ax.set_title("label = %d"%label)
    ax.set_xticks([])
    ax.set_yticks([]) 
plt.show()

在这里插入图片描述

tf.keras.backend.clear_session() #清空会话
#这次我们用函数式构建模型
inputs = layers.Input(shape=(32,32,3))
x = layers.Conv2D(32,kernel_size=(3,3))(inputs)
x = layers.MaxPool2D()(x)
x = layers.Conv2D(64,kernel_size=(5,5))(x)
x = layers.MaxPool2D()(x)
x = layers.Dropout(rate=0.1)(x)
x = layers.Flatten()(x)
x = layers.Dense(32,activation='relu')(x)
outputs = layers.Dense(1,activation = 'sigmoid')(x)

model = models.Model(inputs = inputs,outputs = outputs)

model.summary()
'''
Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 32, 32, 3)]       0         
_________________________________________________________________
conv2d (Conv2D)              (None, 30, 30, 32)        896       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 15, 15, 32)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 11, 11, 64)        51264     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         
_________________________________________________________________
dropout (Dropout)            (None, 5, 5, 64)          0         
_________________________________________________________________
flatten (Flatten)            (None, 1600)              0         
_________________________________________________________________
dense (Dense)                (None, 32)                51232     
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 33        
=================================================================
Total params: 103,425
Trainable params: 103,425
Non-trainable params: 0
_________________________________________________________________
'''

#训练模型通常有3种方法，内置fit方法，内置train_on_batch方法，以及自定义训练循环。此处我们选择最常用也最简单的内置fit方法。
tensorboard_callback = tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)

model.compile(
        optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
        loss=tf.keras.losses.binary_crossentropy,
        metrics=["accuracy"]
    )

history = model.fit(ds_train,epochs= 10,validation_data=ds_test,
                    callbacks = [tensorboard_callback],workers = 4)


'''
Train for 100 steps, validate for 20 steps
Epoch 1/10
100/100 [==============================] - 16s 156ms/step - loss: 0.4830 - accuracy: 0.7697 - val_loss: 0.3396 - val_accuracy: 0.8475
Epoch 2/10
100/100 [==============================] - 14s 142ms/step - loss: 0.3437 - accuracy: 0.8469 - val_loss: 0.2997 - val_accuracy: 0.8680
Epoch 3/10
100/100 [==============================] - 13s 131ms/step - loss: 0.2871 - accuracy: 0.8777 - val_loss: 0.2390 - val_accuracy: 0.9015
Epoch 4/10
100/100 [==============================] - 12s 117ms/step - loss: 0.2410 - accuracy: 0.9040 - val_loss: 0.2005 - val_accuracy: 0.9195
Epoch 5/10
100/100 [==============================] - 13s 130ms/step - loss: 0.1992 - accuracy: 0.9213 - val_loss: 0.1949 - val_accuracy: 0.9180
Epoch 6/10
100/100 [==============================] - 14s 136ms/step - loss: 0.1737 - accuracy: 0.9323 - val_loss: 0.1723 - val_accuracy: 0.9275
Epoch 7/10
100/100 [==============================] - 14s 139ms/step - loss: 0.1531 - accuracy: 0.9412 - val_loss: 0.1670 - val_accuracy: 0.9310
Epoch 8/10
100/100 [==============================] - 13s 134ms/step - loss: 0.1299 - accuracy: 0.9525 - val_loss: 0.1553 - val_accuracy: 0.9340
Epoch 9/10
100/100 [==============================] - 14s 137ms/step - loss: 0.1158 - accuracy: 0.9556 - val_loss: 0.1581 - val_accuracy: 0.9340
Epoch 10/10
100/100 [==============================] - 14s 142ms/step - loss: 0.1006 - accuracy: 0.9617 - val_loss: 0.1614 - val_accuracy: 0.9345
'''

#可以使用evaluate对数据进行评估
val_loss,val_accuracy = model.evaluate(ds_test,workers=4)
print(val_loss,val_accuracy)

#0.16139143370091916 0.9345

可以使用model.predict(ds_test)进行预测。

也可以使用model.predict_on_batch(x_test)对一个批量进行预测。

model.predict(ds_test)
'''
array([[9.9996173e-01],
       [9.5104784e-01],
       [2.8648047e-04],
       ...,
       [1.1484033e-03],
       [3.5589080e-02],
       [9.8537153e-01]], dtype=float32)
'''

for x,y in ds_test.take(1):
    print(model.predict_on_batch(x[0:20]))

'''
tf.Tensor(
[[3.8065155e-05]
 [8.8236779e-01]
 [9.1433197e-01]
 [9.9921846e-01]
 [6.4052093e-01]
 [4.9970779e-03]
 [2.6735585e-04]
 [9.9842811e-01]
 [7.9198682e-01]
 [7.4823302e-01]
 [8.7208226e-03]
 [9.3951421e-03]
 [9.9790359e-01]
 [9.9998581e-01]
 [2.1642199e-05]
 [1.7915063e-02]
 [2.5839690e-02]
 [9.7538447e-01]
 [9.7393811e-01]
 [9.7333014e-01]], shape=(20, 1), dtype=float32)
'''

# 保存权重，该方式仅仅保存权重张量
model.save_weights('./data/tf_model_weights.ckpt',save_format = "tf")

# 保存模型结构与模型参数到文件,该方式保存的模型具有跨平台性便于部署

model.save('./data/tf_model_savedmodel', save_format="tf")
print('export saved model.')