深度卷积网络：实例探究-week2编程题1（Keras 入门 - 笑脸识别）

在完成作业之前需要在虚拟环境中安装TensorFlow和Keras

Keras中几个函数用法https://blog.csdn.net/u012969412/article/details/70882296/

导包

import numpy as np
from keras import layers
from keras.layers import Input, Dense, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D
from keras.layers import AveragePooling2D, MaxPooling2D, Dropout, GlobalMaxPooling2D, GlobalAveragePooling2D
from keras.models import Model
from keras.preprocessing import image
from keras.utils import layer_utils
from keras.utils.data_utils import get_file
from keras.applications.imagenet_utils import preprocess_input
import pydot
from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot
from keras.utils import plot_model
from kt_utils import *

import keras.backend as K
K.set_image_data_format('channels_last')
import matplotlib.pyplot as plt
from matplotlib.pyplot import imshow

加载数据

X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = load_dataset()

# Normalize image vectors
X_train = X_train_orig/255.          #(600, 64, 64, 3)
X_test = X_test_orig/255.            #(150, 64, 64, 3)

# Reshape
Y_train = Y_train_orig.T             #(600, 1)
Y_test = Y_test_orig.T               #(150, 1)

构建模型

Keras框架使用的变量名和我们以前使用的numpy和TensorFlow变量不一样。它不是在前向传播的每一步上创建新变量（比如X, Z1, A1, Z2, A2,…）以便于不同层之间的计算。在Keras中，我们使用X覆盖了所有的值，没有保存每一层结果，我们只需要最新的值，唯一例外的就是X_input，我们将它分离出来是因为它是输入的数据，我们要在最后的创建模型那一步中用到。

def HappyModel(input_shape):
    """
    Implementation of the HappyModel.
    
    Arguments:
    input_shape -- shape of the images of the dataset

    Returns:
    model -- a Model() instance in Keras
    """
    
    ### START CODE HERE ###
    # Feel free to use the suggested outline in the text above to get started, and run through the whole
    # exercise (including the later portions of this notebook) once. The come back also try out other
    # network architectures as well. 
    X_input = Input(input_shape)
    
    # Zero-Padding: pads the border of X_input with zeroes
    X = ZeroPadding2D((3, 3))(X_input)

    # CONV -> BN -> RELU Block applied to X
    X = Conv2D(32, (3, 3), strides = (1, 1), name = 'conv0')(X)
    X = BatchNormalization(axis = 3, name = 'bn0')(X)
    X = Activation('relu')(X)
    
    # MAXPOOL
    X = MaxPooling2D((2, 2), name='max_pool')(X)
    
    X = Conv2D(16, (3, 3), strides = (1, 1), name = 'conv1')(X)  
    X = Activation('relu')(X)
    X = MaxPooling2D((2, 2), name='max_pool1')(X)
    
    # FLATTEN X (means convert it to a vector) + FULLYCONNECTED
    X = Flatten()(X)
    X = Dense(1, activation='sigmoid', name='fc')(X)

    # Create model. This creates your Keras model instance, you'll use this instance to train/test the model.
    model = Model(inputs = X_input, outputs = X, name='HappyModel')
    ### END CODE HERE ###
    
    return model

训练、评估模型

#1.Create the model
happy_model=HappyModel(X_train.shape[1:])
#2.Compile the model
happy_model.compile(optimizer = "adam", loss = "binary_crossentropy", metrics = ["accuracy"])
#3.Train the model on train data
happy_model.fit(x =X_train, y = Y_train, epochs = 40, batch_size = 50)
#4.Test the model on test data
preds=happy_model.evaluate(x = X_test, y = Y_test)

print ("Loss = " + str(preds[0]))
print ("Test Accuracy = " + str(preds[1]))

Epoch 1/40
600/600 [==============================] - 14s - loss: 0.7618 - acc: 0.5683
Epoch 2/40
600/600 [==============================] - 12s - loss: 0.4747 - acc: 0.8100
Epoch 3/40
600/600 [==============================] - 11s - loss: 0.3806 - acc: 0.8467
Epoch 4/40
600/600 [==============================] - 12s - loss: 0.2832 - acc: 0.9183
Epoch 5/40
600/600 [==============================] - 13s - loss: 0.2239 - acc: 0.9367
Epoch 6/40
600/600 [==============================] - 13s - loss: 0.1918 - acc: 0.9500
Epoch 7/40
600/600 [==============================] - 12s - loss: 0.1568 - acc: 0.9550
Epoch 8/40
600/600 [==============================] - 13s - loss: 0.1326 - acc: 0.9683
Epoch 9/40
600/600 [==============================] - 13s - loss: 0.1152 - acc: 0.9717
Epoch 10/40
600/600 [==============================] - 13s - loss: 0.0982 - acc: 0.9800
Epoch 11/40
600/600 [==============================] - 12s - loss: 0.0980 - acc: 0.9817
Epoch 12/40
600/600 [==============================] - 12s - loss: 0.0836 - acc: 0.9817
Epoch 13/40
600/600 [==============================] - 12s - loss: 0.0732 - acc: 0.9883
Epoch 14/40
600/600 [==============================] - 12s - loss: 0.0687 - acc: 0.9817
Epoch 15/40
600/600 [==============================] - 12s - loss: 0.0770 - acc: 0.9817
Epoch 16/40
600/600 [==============================] - 12s - loss: 0.0592 - acc: 0.9933
Epoch 17/40
600/600 [==============================] - 11s - loss: 0.0545 - acc: 0.9900
Epoch 18/40
600/600 [==============================] - 11s - loss: 0.0475 - acc: 0.9917
Epoch 19/40
600/600 [==============================] - 11s - loss: 0.0541 - acc: 0.9867
Epoch 20/40
600/600 [==============================] - 11s - loss: 0.0564 - acc: 0.9867
Epoch 21/40
600/600 [==============================] - 11s - loss: 0.0700 - acc: 0.9783
Epoch 22/40
600/600 [==============================] - 11s - loss: 0.0428 - acc: 0.9900
Epoch 23/40
600/600 [==============================] - 12s - loss: 0.0348 - acc: 0.9917
Epoch 24/40
600/600 [==============================] - 11s - loss: 0.0347 - acc: 0.9900
Epoch 25/40
600/600 [==============================] - 11s - loss: 0.0353 - acc: 0.9933
Epoch 26/40
600/600 [==============================] - 12s - loss: 0.0307 - acc: 0.9917
Epoch 27/40
600/600 [==============================] - 12s - loss: 0.0286 - acc: 0.9950
Epoch 28/40
600/600 [==============================] - 11s - loss: 0.0313 - acc: 0.9967
Epoch 29/40
600/600 [==============================] - 11s - loss: 0.0436 - acc: 0.9883
Epoch 30/40
600/600 [==============================] - 11s - loss: 0.0303 - acc: 0.9900
Epoch 31/40
600/600 [==============================] - 12s - loss: 0.0241 - acc: 0.9983
Epoch 32/40
600/600 [==============================] - 11s - loss: 0.0246 - acc: 0.9933
Epoch 33/40
600/600 [==============================] - 11s - loss: 0.0259 - acc: 0.9950
Epoch 34/40
600/600 [==============================] - 12s - loss: 0.0293 - acc: 0.9917
Epoch 35/40
600/600 [==============================] - 12s - loss: 0.0255 - acc: 0.9900
Epoch 36/40
600/600 [==============================] - 12s - loss: 0.0200 - acc: 0.9933
Epoch 37/40
600/600 [==============================] - 12s - loss: 0.0175 - acc: 0.9983
Epoch 38/40
600/600 [==============================] - 11s - loss: 0.0187 - acc: 0.9967
Epoch 39/40
600/600 [==============================] - 11s - loss: 0.0148 - acc: 0.9983
Epoch 40/40
600/600 [==============================] - 11s - loss: 0.0174 - acc: 0.9983
150/150 [==============================] - 1s
Loss = 0.0885118440787
Test Accuracy = 0.973333330949

预测数据

### START CODE HERE ###
img_path = 'images/a.jpg'
### END CODE HERE ###
img = image.load_img(img_path, target_size=(64, 64))
imshow(img)

x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)

print(happy_model.predict(x))

[[ 1.]]



打印出每一层的大小细节

print(happy_model.summary())

_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) (None, 64, 64, 3) 0
_________________________________________________________________
zero_padding2d_1 (ZeroPaddin (None, 70, 70, 3) 0
_________________________________________________________________
conv0 (Conv2D) (None, 68, 68, 32) 896
_________________________________________________________________
bn0 (BatchNormalization) (None, 68, 68, 32) 128
_________________________________________________________________
activation_1 (Activation) (None, 68, 68, 32) 0
_________________________________________________________________
max_pool (MaxPooling2D) (None, 34, 34, 32) 0
_________________________________________________________________
conv1 (Conv2D) (None, 32, 32, 16) 4624
_________________________________________________________________
activation_2 (Activation) (None, 32, 32, 16) 0
_________________________________________________________________
max_pool1 (MaxPooling2D) (None, 16, 16, 16) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 4096) 0
_________________________________________________________________
fc (Dense) (None, 1) 4097
=================================================================
Total params: 9,745
Trainable params: 9,681
Non-trainable params: 64
_________________________________________________________________
None

绘制布局图（下载、安装并配置Graphviz）

pip install pydot-ng & pip install graphviz

plot_model(happy_model, to_file='happy_model.png')
SVG(model_to_dot(happy_model).create(prog='dot', format='svg'))