首先看路径：
data文件夹分为，model文件夹，train文件夹和文件夹，model文件夹存放模型文件，train存放cat和dog的两个文件夹图片，
validation和train一样。config.py存放配置的一些参数，util.py定义AlexNet框架，AlexNet实现模型的训练，其他没有用到。
首先在config.py定义配置文件，都是卷积层和全连接层用到的参数，用字典的形式存储，
代码如下：
""" 配置文件：定义参数 """ import tensorflow as tf learning_rate=1e-4 training_iers=200 batch_size=50 display_step=5 n_classes=2 n_fc1=4096 n_fc2=2048 W_conv={'conv1':tf.Variable(tf.truncated_normal([11,11,3,96],stddev=0.0001)), 'conv2': tf.Variable(tf.truncated_normal([5, 5, 96, 256], stddev=0.01)), 'conv3': tf.Variable(tf.truncated_normal([3, 3, 256, 384], stddev=0.01)), 'conv4': tf.Variable(tf.truncated_normal([3, 3, 384, 384], stddev=0.01)), 'conv5': tf.Variable(tf.truncated_normal([3, 3, 384, 256], stddev=0.01)), 'fc1': tf.Variable(tf.truncated_normal([6*6*256,n_fc1], stddev=0.1)), 'fc2': tf.Variable(tf.truncated_normal([n_fc1, n_fc2], stddev=0.1)), 'fc3': tf.Variable(tf.truncated_normal([n_fc2, n_classes], stddev=0.1))} b_conv={'conv1':tf.Variable(tf.constant(0.0,shape=[96],dtype=tf.float32)), 'conv2':tf.Variable(tf.constant(0.1,shape=[256],dtype=tf.float32)), 'conv3':tf.Variable(tf.constant(0.1,shape=[384],dtype=tf.float32)), 'conv4':tf.Variable(tf.constant(0.1,shape=[384],dtype=tf.float32)), 'conv5':tf.Variable(tf.constant(0.1,shape=[256],dtype=tf.float32)), 'fc1': tf.Variable(tf.constant(0.1,shape=[n_fc1],dtype=tf.float32)), 'fc2': tf.Variable(tf.constant(0.1,shape=[n_fc2],dtype=tf.float32)), 'fc3': tf.Variable(tf.constant(0.0,shape=[n_classes],dtype=tf.float32))}
在util.py定义AlexNet框架
""" 定义 AlexNet框架 """ import os import numpy as np from config import W_conv,b_conv,n_classes,learning_rate import tensorflow as tf """ 构建AlexNet模型 """ """ 对特征图进行归一化 采用标准化处理 """ def batch_norm(inputs,is_training,is_conv_out=True,decay=0.999): scale=tf.Variable(tf.ones([inputs.get_shape()[-1]])) beta=tf.Variable(tf.zeros([inputs.get_shape()[-1]])) pop_mean = tf.Variable(tf.zeros([inputs.get_shape()[-1]]),trainable=False) pop_var = tf.Variable(tf.ones([inputs.get_shape()[-1]]),trainable=False) if is_training: if is_conv_out: batch_mean,batch_var=tf.nn.moments(inputs,[0,1,2]) else: batch_mean, batch_var =tf.nn.moments(inputs, [0]) #滑动平均 train_mean=tf.assign(pop_mean,pop_mean*decay+batch_mean*(1-decay)) train_var = tf.assign(pop_var, pop_var * decay + batch_var * (1 - decay)) with tf.control_dependencies([train_mean,train_var]): return tf.nn.batch_normalization(inputs, batch_mean,batch_var,beta,scale,0.001) else: return tf.nn.batch_normalization(inputs, pop_mean, pop_var, beta, scale, 0.001) def build_cnn(x,y): x_image=tf.reshape(x,[-1,227,227,3]) #卷积层1 conv1=tf.nn.relu(tf.nn.conv2d(x_image,W_conv['conv1'],strides=[1,4,4,1],padding='VALID') +b_conv['conv1']) conv1=batch_norm(conv1,True) #池化层1 pool1=tf.nn.avg_pool(conv1,ksize=[1,3,3,1],strides=[1,2,2,1],padding='VALID') #LRN层1 norm1=tf.nn.lrn(pool1,5,bias=1.0,alpha=0.001/9.0,beta=0.75) #卷积层2 conv2=tf.nn.relu(tf.nn.conv2d(norm1,W_conv['conv2'],strides=[1,1,1,1],padding='SAME') +b_conv['conv2']) conv2 = batch_norm(conv2, True) #池化层2 pool2=tf.nn.avg_pool(conv2,ksize=[1,3,3,1],strides=[1,2,2,1],padding='VALID') #LRN层2 norm2=tf.nn.lrn(pool2,5,bias=1.0,alpha=0.001/9.0,beta=0.75) #卷积层3 conv3=tf.nn.relu(tf.nn.conv2d(norm2,W_conv['conv3'],strides=[1,1,1,1],padding='SAME') +b_conv['conv3']) conv3 = batch_norm(conv3, True) #卷积层4 conv4=tf.nn.relu(tf.nn.conv2d(conv3,W_conv['conv4'],strides=[1,1,1,1],padding='SAME') +b_conv['conv4']) conv4 = batch_norm(conv4, True) #卷积层5 conv5=tf.nn.relu(tf.nn.conv2d(conv4,W_conv['conv5'],strides=[1,1,1,1],padding='SAME') +b_conv['conv5']) conv5 = batch_norm(conv5, True) #池化层5 pool5=tf.nn.avg_pool(conv5,ksize=[1,3,3,1],strides=[1,2,2,1],padding='VALID') #全连接层1 fc1=tf.matmul(tf.reshape(pool5, [-1, 6 * 6 * 256]), W_conv['fc1']) + b_conv['fc1'] fc1 = batch_norm(fc1, True,is_conv_out=False) fc1=tf.nn.relu(fc1) fc1=tf.nn.dropout(fc1,0.5) #全连接层2 fc2=tf.matmul(fc1, W_conv['fc2']) + b_conv['fc2'] fc2 = batch_norm(fc2, True, is_conv_out=False) fc2=tf.nn.relu(fc2) fc2=tf.nn.dropout(fc2,0.5) #全连接层3 fc3=tf.matmul(fc2,W_conv['fc3'])+b_conv['fc3'] #定义损失 loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=fc3,labels=y)) optimizer=tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(loss) #评估模型 accuarcy=tf.reduce_mean(tf.cast(tf.equal(tf.argmax(fc3,1),tf.argmax(y,1)),dtype=tf.float32)) return loss,optimizer,accuarcy """ 读取train.tfrecords并返回image和label 这个是已经做好的.tfrecords文件对于太大的不适合 """ def read_and_decode(filename,batch_size): filename_queue=tf.train.string_input_producer([filename])# 按队列的形式读取 reader=tf.TFRecordReader() _,serialized_example=reader.read(filename_queue)#返回文件名和文件 features=tf.parse_single_example(serialized_example, features={ 'label':tf.FixedLenFeature([],tf.int64),#与存储的类型一致 'image':tf.FixedLenFeature([],tf.string) }) img=tf.decode_raw(features['image'],tf.uint8) img=tf.reshape(img,shape=[227,227,3]) #img = tf.cast(img, dtype=tf.float32) * (1.0 / 128) - 0.5 label = tf.cast(features['label'], dtype=tf.int32) img_batch, label_batch = tf.train.shuffle_batch([img, label], batch_size=batch_size, capacity=64, min_after_dequeue=32, num_threads=64) return img_batch,tf.reshape(label_batch,[batch_size]) """ 将图片的路径和对应的标签存储在list中返回 """ def get_file(dir): images = [] temp = [] for root,dirs,files in os.walk(dir): for name in files: images.append(os.path.join(root,name)) for name in dirs:#以dogs cats文件夹形式读取 temp.append(os.path.join(root,name)) labels=[] for one_folder in temp: n_img=len(os.listdir(one_folder))#展开cats或者dogs的图片 letter=one_folder.split('/')[-1] if letter=='cats': labels=np.append(labels,n_img*[0])#np.append拼接 0是cat 1是dog else: labels=np.append(labels,n_img*[1]) #打乱 temp=np.array([images,labels]) temp=temp.transpose() np.random.shuffle(temp) image_list=list(temp[:,0]) label_list=list(temp[:,1]) label_list=[int(float(i)) for i in label_list] return image_list,label_list """ 太大的TFrecord数据集占内存，用此方法可将需要数量的图片转换成TFrecord即可 """ def get_batch(image_list,label_list,img_width,img_height,batch_size,capacity): image=tf.cast(image_list,dtype=tf.string) label=tf.cast(label_list, dtype=tf.int32) input_queue=tf.train.slice_input_producer([image,label]) label=input_queue[1] image_contents=tf.read_file(input_queue[0]) image=tf.image.decode_jpeg(image_contents,channels=3) image=tf.image.resize_image_with_crop_or_pad(image,img_width,img_height) image=tf.image.per_image_standardization(image)#将图片标准化 image_batch,label_batch=tf.train.batch([image,label],batch_size=batch_size,num_threads=64,capacity=capacity) label_batch=tf.reshape(label_batch,[batch_size]) return image_batch,label_batch """ 实现标签的one-hot """ # #转换one-hot def one_hot(label): n_classes=max(label)+1 label = np.eye(n_classes)[label.reshape(-1)] return label # 转换one-hot # def one_hot(labels): # n_samples=len(labels) # n_class=n_samples+1 # onehot_label=np.zeros((n_samples,n_class)) # onehot_label[np.arange(n_samples),labels]=1 # return onehot_label
最后在AlexNet.py调用即可
import tensorflow as tf import util import config import time import matplotlib.pyplot as plt """ 直接读取整个train.tfrecord """ #filename='./data/train/train.tfrecords' #img_batch,label_batch=util.read_and_decode(filename,batch_size=32) """ 按照需求来读取图片 """ dir='./data/train' image_list,label_list=util.get_file(dir) img_batch,label_batch=util.get_batch(image_list,label_list,227,227,batch_size=32,capacity=64) x=tf.placeholder(shape=[None,227,227,3],dtype=tf.float32) y=tf.placeholder(shape=[None,config.n_classes],dtype=tf.float32) loss,optimizer,accuarcy=util.build_cnn(x,y) def train(epoch): init=tf.global_variables_initializer() saver = tf.train.Saver() with tf.Session() as sess: sess.run(init) costs=[] start_time=time.time() save_model='./data/model/AlexNetModel.ckpt' train_writer=tf.summary.FileWriter('./log',sess.graph) #启动线程 #coord=tf.train.Coordinator() threads = tf.train.start_queue_runners(sess=sess) for i in range(epoch): image,label=sess.run([img_batch,label_batch]) labels = util.one_hot(label) _,cost,train_accuarcy=sess.run([optimizer,loss,accuarcy],feed_dict={x:image,y:labels}) print('step={},loss={},train_accuarcy={}'.format(i,cost,train_accuarcy)) costs.append(cost) end_time=time.time() print('step={},time={}'.format(i,(end_time-start_time))) print('') print('optimization is finish') saver.save(sess,save_model) print('model save finished') # coord.request_stop() # coord.join(threads) plt.plot(costs) plt.show() plt.xlabel('iter') plt.ylabel('cost') if __name__ == '__main__': train(epoch=50)由于做好的train.TFrecorder数据集太大，后面改成按照所需的数据量转换成TFrecorder。