tensorflow mnist手写数字（三）从零开始

看完官方的教程，发现官方的数据都已经做好了，现在我们有自己的手写数字图片，也就是从0开始，只有手写数字图片，该怎么识别呢？

感谢这位博主  http://www.cnblogs.com/denny402/p/5684431.html

mnist图片数据下载：http://pan.baidu.com/s/1pLMV4Kz

首先建立模型文件，我们命名为mnist_inference.py

代码如下

#!/usr/bin/env python2# -*- coding: utf-8 -*-"""Created on Mon Oct 23 16:31:46 2017@author: hjxu"""#!/usr/bin/env python3# -*- coding: utf-8 -*-import tensorflow as tf# 定义神经网络相关的参数w = 32h = 32c = 1# 定义神经网络的前向传播过程。# 这里添加了一个新的参数train，用于区别训练过程和测试过程。# 在这个程序中将用到dropout方法，dropout可以进一步提升模型可靠性并防止过拟合，dropout过程只在训练时使用。def inference(input_tensor, train, regularizer):        #第一层：卷积层，过滤器的尺寸为5×5，深度为6,不使用全0补充，步长为1。    #尺寸变化：32×32×1->28×28×6    with tf.variable_scope('layer1-conv1'):        conv1_weights = tf.get_variable('weight',[5,5,c,6],initializer=tf.truncated_normal_initializer(stddev=0.1))        conv1_biases = tf.get_variable('bias',[6],initializer=tf.constant_initializer(0.0))        conv1 = tf.nn.conv2d(input_tensor,conv1_weights,strides=[1,1,1,1],padding='VALID')        relu1 = tf.nn.relu(tf.nn.bias_add(conv1,conv1_biases))    #第二层：池化层，过滤器的尺寸为2×2，使用全0补充，步长为2。    #尺寸变化：28×28×6->14×14×6    with tf.name_scope('layer2-pool1'):        pool1 = tf.nn.max_pool(relu1,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')    #第三层：卷积层，过滤器的尺寸为5×5，深度为16,不使用全0补充，步长为1。    #尺寸变化：14×14×6->10×10×16    with tf.variable_scope('layer3-conv2'):        conv2_weights = tf.get_variable('weight',[5,5,6,16],initializer=tf.truncated_normal_initializer(stddev=0.1))        conv2_biases = tf.get_variable('bias',[16],initializer=tf.constant_initializer(0.0))        conv2 = tf.nn.conv2d(pool1,conv2_weights,strides=[1,1,1,1],padding='VALID')        relu2 = tf.nn.relu(tf.nn.bias_add(conv2,conv2_biases))    #第四层：池化层，过滤器的尺寸为2×2，使用全0补充，步长为2。    #尺寸变化：10×10×6->5×5×16    with tf.variable_scope('layer4-pool2'):        pool2 = tf.nn.max_pool(relu2,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')    #将第四层池化层的输出转化为第五层全连接层的输入格式。第四层的输出为5×5×16的矩阵，然而第五层全连接层需要的输入格式    #为向量，所以我们需要把代表每张图片的尺寸为5×5×16的矩阵拉直成一个长度为5×5×16的向量。    #举例说，每次训练64张图片，那么第四层池化层的输出的size为(64,5,5,16),拉直为向量，nodes=5×5×16=400,尺寸size变为(64,400)    pool_shape = pool2.get_shape().as_list()    nodes = pool_shape[1]*pool_shape[2]*pool_shape[3]    reshaped = tf.reshape(pool2,[-1,nodes])    #第五层：全连接层，nodes=5×5×16=400，400->120的全连接    #尺寸变化：比如一组训练样本为64，那么尺寸变化为64×400->64×120    #训练时，引入dropout，dropout在训练时会随机将部分节点的输出改为0，dropout可以避免过拟合问题。    #这和模型越简单越不容易过拟合思想一致，和正则化限制权重的大小，使得模型不能任意拟合训练数据中的随机噪声，以此达到避免过拟合思想一致。    #本文最后训练时没有采用dropout，dropout项传入参数设置成了False，因为训练和测试写在了一起没有分离，不过大家可以尝试。    with tf.variable_scope('layer5-fc1'):        fc1_weights = tf.get_variable('weight',[nodes,120],initializer=tf.truncated_normal_initializer(stddev=0.1))        if regularizer != None:            tf.add_to_collection('losses',regularizer(fc1_weights))        fc1_biases = tf.get_variable('bias',[120],initializer=tf.constant_initializer(0.1))        fc1 = tf.nn.relu(tf.matmul(reshaped,fc1_weights) + fc1_biases)        if train:            fc1 = tf.nn.dropout(fc1,0.5)    #第六层：全连接层，120->84的全连接    #尺寸变化：比如一组训练样本为64，那么尺寸变化为64×120->64×84    with tf.variable_scope('layer6-fc2'):        fc2_weights = tf.get_variable('weight',[120,84],initializer=tf.truncated_normal_initializer(stddev=0.1))        if regularizer != None:            tf.add_to_collection('losses',regularizer(fc2_weights))        fc2_biases = tf.get_variable('bias',[84],initializer=tf.truncated_normal_initializer(stddev=0.1))        fc2 = tf.nn.relu(tf.matmul(fc1,fc2_weights) + fc2_biases)        if train:            fc2 = tf.nn.dropout(fc2,0.5)    #第七层：全连接层（近似表示），84->10的全连接    #尺寸变化：比如一组训练样本为64，那么尺寸变化为64×84->64×10。最后，64×10的矩阵经过softmax之后就得出了64张图片分类于每种数字的概率，    #即得到最后的分类结果。    with tf.variable_scope('layer7-fc3'):        fc3_weights = tf.get_variable('weight',[84,10],initializer=tf.truncated_normal_initializer(stddev=0.1))        if regularizer != None:            tf.add_to_collection('losses',regularizer(fc3_weights))        fc3_biases = tf.get_variable('bias',[10],initializer=tf.truncated_normal_initializer(stddev=0.1))        logit = tf.matmul(fc2,fc3_weights) + fc3_biases        tf.add_to_collection('logits',logit)    return logit

网络结构定义好了，要再编写一个训练的脚本，命名为mnist_train.py

#!/usr/bin/env python2# -*- coding: utf-8 -*-"""Created on Mon Oct 23 16:32:54 2017@author: hjxu"""import osimport numpy as npimport tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_data# 加载mnist_inference.py中定义的常量和前向传播的函数from mnist_inference import  inferencefrom skimage import io,transformw = 32h = 32c = 1###load_file返回三个值，第一个是txt文件的每行空格前的字符，对应路径，第二个是每行的label，第三个是总个数###def load_file(examples_list_file):    # type: (object) -> object    lines = np.genfromtxt(examples_list_file, delimiter=" ", dtype=[('col1', 'S120'), ('col2', 'i8')])    examples = []    labels = []    for example, label in lines:        examples.append(example)        labels.append(label)    return np.asarray(examples), np.asarray(labels), len(lines)    ###load_file返回三个值，第一个是txt文件的每行空格前的字符，对应路径，第二个是每行的label，第三个是总个数######  测试load——file###的一个例子###def test_load_file():    train_file = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/mnist_data/train/train.txt'    examples, labels, examples_num = load_file(train_file)    for i in range(examples_num):        print 'No',i,'path',examples[i],'label is',labels[i]###测试load——file###的一个例子###def read_img(img_folder,path_txt):    images, labels, examples_num = load_file(path_txt)    img_result = []    label_result = []    for i in range(examples_num):        image = io.imread(images[i])        image = transform.resize(image,(w,h,c))        img_result.append(image)        label_result.append(labels[i])    return np.asarray(img_result,dtype=np.float32),np.asarray(label_result,dtype=np.int32)        img_folder = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/'#存放图片的根目录train_path_txt =  '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/mnist/train/train.txt'#图片列表的绝对路径test_path_txt =  '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/mnist/test/test.txt'     logs_train_dir = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/ckpt/' #模型日志保存的文件夹路径train_data,train_label = read_img(img_folder,train_path_txt)test_data,test_label = read_img(img_folder,test_path_txt)#打乱训练数据及测试数据train_image_num = len(train_data)train_image_index = np.arange(train_image_num)np.random.shuffle(train_image_index)train_data = train_data[train_image_index]train_label = train_label[train_image_index]test_image_num = len(test_data)test_image_index = np.arange(test_image_num)np.random.shuffle(test_image_index)test_data = test_data[test_image_index]test_label = test_label[test_image_index]#搭建CNNx = tf.placeholder(tf.float32,[None,w,h,c],name='x')y_ = tf.placeholder(tf.int32,[None],name='y_')regularizer = tf.contrib.layers.l2_regularizer(0.001)y = inference(x,False,regularizer)cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y,labels=y_)cross_entropy_mean = tf.reduce_mean(cross_entropy)loss = cross_entropy_mean + tf.add_n(tf.get_collection('losses'))train_op = tf.train.AdamOptimizer(0.001).minimize(loss)correct_prediction = tf.equal(tf.cast(tf.argmax(y,1),tf.int32),y_)accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))#每次获取batch_size个样本进行训练或测试def get_batch(data,label,batch_size):    for start_index in range(0,len(data)-batch_size+1,batch_size):        slice_index = slice(start_index,start_index+batch_size)        yield data[slice_index],label[slice_index]#创建Session会话with tf.Session() as sess:    #初始化所有变量(权值，偏置等)    sess.run(tf.global_variables_initializer())    #将所有样本训练10次，每次训练中以64个为一组训练完所有样本。    #train_num可以设置大一些。    train_num = 40    batch_size = 64    train_writer = tf.summary.FileWriter(logs_train_dir, sess.graph)    saver = tf.train.Saver()    for i in range(train_num):        train_loss,train_acc,batch_num = 0, 0, 0        for train_data_batch,train_label_batch in get_batch(train_data,train_label,batch_size):            _,err,acc = sess.run([train_op,loss,accuracy],feed_dict={x:train_data_batch,y_:train_label_batch})            train_loss+=err;train_acc+=acc;batch_num+=1        print("train loss:",train_loss/batch_num)        print("train acc:",train_acc/batch_num)        test_loss,test_acc,batch_num = 0, 0, 0        for test_data_batch,test_label_batch in get_batch(test_data,test_label,batch_size):            err,acc = sess.run([loss,accuracy],feed_dict={x:test_data_batch,y_:test_label_batch})            test_loss+=err;test_acc+=acc;batch_num+=1        print("test loss:",test_loss/batch_num)        print("test acc:",test_acc/batch_num)        if i % 20 == 0 or (i + 1) == train_num:            checkpoint_path = os.path.join(logs_train_dir, 'model.ckpt')            saver.save(sess, checkpoint_path, global_step=i+1)

在ckpt文件夹中就会出现一些保存模型的文件，然后我们现在需要测试一张图片来测试我们的模型

我们建议一个脚本命名evaluation_one_image.py

# -*- coding: utf-8 -*-import tensorflow as tffrom sys import pathimport numpy as npimport tensorflow as tfimport timeimport cv2from PIL import Imageimport randomimport linecacheimport matplotlib.pyplot as plt from mnist_inference import  inferencew = 32h = 32c = 1##############测试一张指定的图片#############33#image_root = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/mnist/test/6/00011.png'#im = Image.open(image_root)#plt.imshow(im)#plt.show()#im = im.resize((32,32))##############测试一张指定的图片#############33###########随机从txt文件中测试一张图片##################3def load_file(examples_list_file):    # type: (object) -> object    lines = np.genfromtxt(examples_list_file, delimiter=" ", dtype=[('col1', 'S120'), ('col2', 'i8')])    examples = []    labels = []    for example, label in lines:        examples.append(example)        labels.append(label)    return np.asarray(examples), np.asarray(labels), len(lines)    def get_one_img(examples_list_file):    img,label,num = load_file(examples_list_file)    i = random.randint(1, (num))    print img[i]    return img[i]############3随机从txt文件中测试一张图片##################3examples_list_file = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/mnist/train/train.txt'image_root = '/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/'+str(get_one_img(examples_list_file))im = Image.open(image_root)plt.imshow(im)plt.show()im = im.resize((32,32))im = np.array(im).astype(np.float32)im = np.reshape(im, [-1, 32 * 32 * 1])batch_xs = np.reshape(im, [-1, 32, 32, 1])x = tf.placeholder(tf.float32, shape=[1,32, 32, 1])##logit = inference(x, False, None)with tf.Session() as sess:  new_saver = tf.train.import_meta_graph('/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/ckpt/model.ckpt-21.meta')  new_saver.restore(sess,  tf.train.latest_checkpoint('/home/hjxu/PycharmProjects/tf_examples/mnist_to_jjh/ckpt/'))      # tf.get_collection() 返回一个list. 但是这里只要第一个参数即可  logit = tf.nn.softmax(tf.get_collection('logits')[0])  graph = tf.get_default_graph()  # 因为y中有placeholder，所以sess.run(y)的时候还需要用实际待预测的样本以及相应的参数来填充这些placeholder，而这些需要通过graph的get_operation_by_name方法来获取。  input_x = graph.get_operation_by_name('x').outputs[0]  out = sess.run(logit, feed_dict={input_x:batch_xs})  print np.argmax(out)

这个脚本可以指定测试一张图片，也可以随机读取一张图片