吴恩达作业10：用卷积神经网络识别人脸happy（基于Keras）

数据集提供的代码放在kt_utils.py：

import keras.backend as K
import math
import numpy as np
import h5py
import matplotlib.pyplot as plt

def mean_pred(y_true, y_pred):
    return K.mean(y_pred)

def load_dataset():
    train_dataset = h5py.File('datasets/train_happy.h5', "r")
    train_set_x_orig = np.array(train_dataset["train_set_x"][:]) # your train set features
    train_set_y_orig = np.array(train_dataset["train_set_y"][:]) # your train set labels
    #print(train_set_x_orig.shape) ##(600, 64, 64, 3)
    #print(train_set_y_orig)##(600,)
    test_dataset = h5py.File('datasets/test_happy.h5', "r")
    test_set_x_orig = np.array(test_dataset["test_set_x"][:]) # your test set features
    test_set_y_orig = np.array(test_dataset["test_set_y"][:]) # your test set labels
    #print(test_set_x_orig.shape)#(150, 64, 64, 3)
    #print(test_set_y_orig.shape)#(150,)
    classes = np.array(test_dataset["list_classes"][:]) # the list of classes
    #print(classes) #[0,1]
    train_set_y_orig = train_set_y_orig.reshape((1, train_set_y_orig.shape[0]))
    test_set_y_orig = test_set_y_orig.reshape((1, test_set_y_orig.shape[0]))
    #print(train_set_y_orig.shape)  (1, 600)
    #print(test_set_y_orig.shape)   (1, 150)
    return train_set_x_orig, train_set_y_orig, test_set_x_orig, test_set_y_orig, classes
#load_dataset()

查看数据集：

import kt_utils
import cv2
import matplotlib.pyplot as plt
train_set_x_orig, train_set_Y, test_set_x_orig, test_set_Y, classes = kt_utils.load_dataset()
print('训练样本={}'.format(train_set_x_orig.shape))
print('训练样本标签={}'.format(train_set_Y.shape))
print('测试样本={}'.format(test_set_x_orig.shape))
print('测试样本标签={}'.format(test_set_Y.shape))
print('第五个样本={}'.format(train_set_Y[0,5]))
cv2.imshow('1.jpg',train_set_x_orig[5,:,:,:])
cv2.waitKey()
print('第六个样本={}'.format(train_set_Y[0,6]))
cv2.imshow('1.jpg',train_set_x_orig[6,:,:,:])
cv2.waitKey()
# plt.imshow(train_set_x_orig[5,:,:,:])
# plt.show()

打印结果：可看出600个训练样本，150个测试样本，size=（64，64，3）,其中happy的标签为1，not happy的标签为0，故标签也要经过one-hot。

开始训练模型代码如下：

import numpy as np
import matplotlib.pyplot as plt
from keras.models import Model
from keras.layers import Input,ZeroPadding2D,Conv2D,BatchNormalization,Activation,MaxPooling2D
from keras.layers import Flatten,Dense
import kt_utils
from keras.preprocessing import image
from keras.applications.imagenet_utils import preprocess_input
from keras.utils import plot_model
from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot
import time
import cv2
"""
转换成one-hot
"""
def convert_to_one_hot(Y, C):
    Y = np.eye(C)[Y.reshape(-1)].T
    return Y
"""
获取数据  并将标签转换成one-hot
"""
def convert_data():
    train_set_x_orig, train_set_y_orig, test_set_x_orig, test_set_y_orig, classes=kt_utils.load_dataset()
    train_x=train_set_x_orig/255
    test_x = test_set_x_orig / 255
    train_y=convert_to_one_hot(train_set_y_orig,2).T
    test_y = convert_to_one_hot(test_set_y_orig, 2).T
    #print(train_y.shape)
    return train_x,train_y,test_x,test_y
"""
查看样本
"""
def test():
    train_set_x_orig, train_set_y_orig, test_set_x_orig, test_set_y_orig, classes=kt_utils.load_dataset()
    X_train=train_set_x_orig/255
    X_test = test_set_x_orig / 255
    Y_train=train_set_y_orig.T
    Y_test = test_set_y_orig.T
    plt.imshow(train_set_x_orig[63,:,:,:])
    plt.show()
"""
构建CNN模型
"""
def model(input_shape):
    X_input=Input(input_shape)
    print('输入尺寸={}'.format(X_input.shape))
    #Zero padding
    X=ZeroPadding2D((3,3))(X_input)
    print('输补完零尺寸={}'.format(X.shape))
    #CONV->BN->RELU
    X=Conv2D(32,(7,7),strides=(1,1),name='conv0')(X)
    print('第一次卷积尺寸={}'.format(X.shape))
    X=BatchNormalization(axis=-1,name='bn0')(X)
    X=Activation('relu')(X)
    #MAXPOOL
    X=MaxPooling2D((2,2),name='max_pool')(X)
    print('第一池化尺寸={}'.format(X.shape))
    #FLATTEN+FULLYCONNECTED
    X=Flatten()(X)
    X=Dense(2,activation='sigmoid',name='fc')(X)
    model=Model(inputs=X_input,outputs=X,name='HappyModel')
    return model
"""
测试模型
"""
def testModel():
    train_x, train_y, test_x, test_y=convert_data()
    #定义好模型结构
    happyModel=model(input_shape=[64,64,3])
    #模型编译
    happyModel.compile(optimizer='adam',loss='binary_crossentropy',metrics=['accuracy'])
    #模型训练
    start_time=time.time()
    print('============模型开始训练=====================')
    happyModel.fit(x=train_x,y=train_y,epochs=1,batch_size=32)
    end_time=time.time()
    print('train_time={}'.format(end_time-start_time))
    # save the model
    #happyModel.save('my_model_v1.h5')
    print('============模型开始测试=====================')
    preds=happyModel.evaluate(x=test_x,y=test_y,batch_size=32)
    print()
    print('loss={}'.format(preds[0]))
    print('Test accuarcy={}'.format(preds[1]))
    #打印参数
    happyModel.summary()
    #可视化模型
    plot_model(happyModel,to_file='HappyModel.png')
    SVG(model_to_dot(happyModel).create(prog='dot',format='svg'))
    #test my_image
    print('============测试自己的照片===================')
    path = 'images/my_image.jpg'
    img = image.load_img(path, target_size=(64, 64))
    plt.imshow(img)
    plt.show()
    x = image.img_to_array(img)  # (64,64,3)
    x = x.reshape(1, 64, 64, 3)
    x=preprocess_input(x)
    y=happyModel.predict(x)
    print('预测值={}'.format(y))
def testPicture():
    # img=cv2.imread('my_image.jpg')
    # cv2.imshow('img',img)
    path='images/my_image.jpg'
    img=image.load_img(path,target_size=(64,64))
    plt.imshow(img)
    plt.show()
    print('img=',img)

    x=image.img_to_array(img)#(64,64,3)
    print('x=',x)
    print(x.shape)
    #x = np.expand_dims(x, axis=0)#(1,64,64,3)
    x=x.reshape(64,64,3)
    print('x=',x)
    print(x.shape)
    plt.imshow(x)
    plt.show()
if __name__=='__main__':
    #test()
    testModel()
    #testPicture()

打印结果：？号代表样本数，可知池化过后尺寸为（32，32，32）

训练次数是一次，结果如下：可知每张图片训练时间为17ms，600张时间为10s，跟自己记录的train_time差一点点，因为还有别的开支。

测试结果：测试精度一般

打印出模型参数，并且可视化：可看成conv0参数=7×7×3×32（W的参数）+32（b的参数）=4736

测试自己的照片：

打印结果：貌似是happy 貌似又不是 样本量少 加上训练次数少 肯定是不准的。