cifar数据集的预处理和训练

注:以下代码从mooc网<人工智能实践：Tensorflow笔记>课程作业给出参考代码和讨论区同学回复代码学习修改得到,侵权立删.

网址:https://www.icourse163.org/learn/PKU-1002536002?tid=1206591210#/learn/forumdetail?pid=1212984121

forward代码:

#coding:utf-8
import tensorflow as tf
IMAGE_SIZE = 32 #input:32*32
NUM_CHANNELS = 3 #color:3 
CONV1_SIZE = 5
CONV1_KERNEL_NUM = 32
CONV2_SIZE = 5
CONV2_KERNEL_NUM = 64
FC_SIZE = 512
OUTPUT_NODE = 10

def get_weight(shape, regularizer):
	w = tf.Variable(tf.truncated_normal(shape,stddev=0.1))
	if regularizer != None: tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w)) 
	return w

def get_bias(shape): 
	b = tf.Variable(tf.zeros(shape))  
	return b

def conv2d(x,w):  
	return tf.nn.conv2d(x, w, strides=[1, 1, 1, 1], padding='SAME')

def max_pool_2x2(x):  
	return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') 

def forward(x, train, regularizer):
    conv1_w = get_weight([CONV1_SIZE, CONV1_SIZE, NUM_CHANNELS, CONV1_KERNEL_NUM], regularizer) 
    conv1_b = get_bias([CONV1_KERNEL_NUM]) 
    conv1 = conv2d(x, conv1_w) 
    relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_b)) 
    pool1 = max_pool_2x2(relu1) 

    conv2_w = get_weight([CONV2_SIZE, CONV2_SIZE, CONV1_KERNEL_NUM, CONV2_KERNEL_NUM],regularizer) 
    conv2_b = get_bias([CONV2_KERNEL_NUM])
    conv2 = conv2d(pool1, conv2_w) 
    relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_b))
    pool2 = max_pool_2x2(relu2)

    pool_shape = pool2.get_shape().as_list() 
    nodes = pool_shape[1] * pool_shape[2] * pool_shape[3] 
    reshaped = tf.reshape(pool2, [pool_shape[0], nodes]) 

    fc1_w = get_weight([nodes, FC_SIZE], regularizer) 
    fc1_b = get_bias([FC_SIZE]) 
    fc1 = tf.nn.relu(tf.matmul(reshaped, fc1_w) + fc1_b) 
    if train: fc1 = tf.nn.dropout(fc1, 0.5)

    fc2_w = get_weight([FC_SIZE, OUTPUT_NODE], regularizer)
    fc2_b = get_bias([OUTPUT_NODE])
    y = tf.matmul(fc1, fc2_w) + fc2_b
    return y 

backward代码:

#coding:utf-8
import tensorflow as tf
import cifar_lenet5_forward
import os
import numpy as np
from read_data import DataLoad

BATCH_SIZE = 100
LEARNING_RATE_BASE =  0.005 
LEARNING_RATE_DECAY = 0.99 
REGULARIZER = 0.0001 
STEPS = 5000
MOVING_AVERAGE_DECAY = 0.99 
MODEL_SAVE_PATH="./model/" 
MODEL_NAME="cifar_model" 

def backward(cifar):
    x = tf.placeholder(tf.float32,[
	BATCH_SIZE,
	cifar_lenet5_forward.IMAGE_SIZE,
	cifar_lenet5_forward.IMAGE_SIZE,
	cifar_lenet5_forward.NUM_CHANNELS]) 
    y_ = tf.placeholder(tf.float32, [None, cifar_lenet5_forward.OUTPUT_NODE])
    y = cifar_lenet5_forward.forward(x,True, REGULARIZER) 
    global_step = tf.Variable(0, trainable=False) 

    ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))
    cem = tf.reduce_mean(ce) 
    loss = cem + tf.add_n(tf.get_collection('losses')) 

    learning_rate = tf.train.exponential_decay( 
        LEARNING_RATE_BASE,
        global_step,
        cifar.num_examples / BATCH_SIZE, 
		LEARNING_RATE_DECAY,
        staircase=True) 
    
    train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)

    ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
    ema_op = ema.apply(tf.trainable_variables())
    with tf.control_dependencies([train_step, ema_op]): 
        train_op = tf.no_op(name='train')

    saver = tf.train.Saver() 

    with tf.Session() as sess: 
        init_op = tf.global_variables_initializer() 
        sess.run(init_op) 

        ckpt = tf.train.get_checkpoint_state(MODEL_SAVE_PATH) 
        if ckpt and ckpt.model_checkpoint_path:
        	saver.restore(sess, ckpt.model_checkpoint_path) 

        for i in range(STEPS):
            xs, ys = cifar.next_batch(BATCH_SIZE)
            reshaped_xs = np.reshape(xs,(  
		    BATCH_SIZE,
        	cifar_lenet5_forward.IMAGE_SIZE,
        	cifar_lenet5_forward.IMAGE_SIZE,
        	cifar_lenet5_forward.NUM_CHANNELS))
            _, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: reshaped_xs, y_: ys}) 
            if i % 100 == 0: 
                print("After %d training step(s), loss on training batch is %g." % (step, loss_value))
                saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step)
'''
def main():
    cifar = input_data.read_data_sets("./data/", one_hot=True) 
    backward(cifar)
'''

def main():
    #cifar10 = input_data.read_data_sets("./data/", one_hot=True)  # 读入 Cifar-10 数据
    cifar10_train = DataLoad("./train")
    backward(cifar10_train)

if __name__ == '__main__':
    main()

test代码:

#coding:utf-8
import time
import tensorflow as tf
import cifar_lenet5_forward
import cifar_lenet5_backward
import numpy as np
from read_data import DataLoad

TEST_INTERVAL_SECS = 5

def test(cifar):
    with tf.Graph().as_default() as g: 
        x = tf.placeholder(tf.float32,[
            cifar.num_examples,
            cifar_lenet5_forward.IMAGE_SIZE,
            cifar_lenet5_forward.IMAGE_SIZE,
            cifar_lenet5_forward.NUM_CHANNELS]) 
        y_ = tf.placeholder(tf.float32, [None, cifar_lenet5_forward.OUTPUT_NODE])
        y = cifar_lenet5_forward.forward(x,False,None)

        ema = tf.train.ExponentialMovingAverage(cifar_lenet5_backward.MOVING_AVERAGE_DECAY)
        ema_restore = ema.variables_to_restore()
        saver = tf.train.Saver(ema_restore)
		
        correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1)) 
        accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 

        while True:
            with tf.Session() as sess:
                ckpt = tf.train.get_checkpoint_state(cifar_lenet5_backward.MODEL_SAVE_PATH)
                if ckpt and ckpt.model_checkpoint_path:
                    saver.restore(sess, ckpt.model_checkpoint_path)
					
                    global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1] 
                    reshaped_x = np.reshape(cifar.xs,(
                    cifar.num_examples,
        	        cifar_lenet5_forward.IMAGE_SIZE,
        	        cifar_lenet5_forward.IMAGE_SIZE,
        	        cifar_lenet5_forward.NUM_CHANNELS))
                    accuracy_score = sess.run(accuracy, feed_dict={x:reshaped_x,y_:cifar.ys}) 
                    print("After %s training step(s), test accuracy = %g" % (global_step, accuracy_score))
                else:
                    print('No checkpoint file found')
                    return
            time.sleep(TEST_INTERVAL_SECS) 
'''
def main():
    cifar = input_data.read_data_sets("./data/", one_hot=True)
    test(cifar)
'''

def main():
    #cifar10 = input_data.read_data_sets("./data/", one_hot=True)
    cifar10_test = DataLoad("./test")
    test(cifar10_test)

if __name__ == '__main__':
    main()

read_data代码:

#coding:utf-8
from PIL import Image
import numpy as np
import os
import random
class DataLoad:
    def __init__(self,data_path):
        '''
        初始化加载 数据目录
        :param data_path: 包含图片数据集的目录，内部应该包含不同分类的的图片，每种类型是一个子目录
        '''
        self.xs = []
        self.ys = []
        # 还是先写死吧，简单点
        lables_d = {
            0: "airplane",
            1: "automobile",
            2: "bird",
            3: "cat",
            4: "deer",
            5: "dog",
            6: "frog",
            7: "horse",
            8: "ship",
            9: "truck"
        }
        a_tmp_list = []
        for k,v in lables_d.items():
            a_path = data_path + "/" + v
            for file in os.listdir(a_path):
                x = a_path + "/" + file # pre_pic()
                a_tmp_list.append((x, k))  # 这里，暂不读取文件.记录每个文件名对应的label
                #print(x,k)
        #
        random.shuffle(a_tmp_list) # 对列表随机打乱
        #
        for (file_path, k) in a_tmp_list:
            y = np.zeros(10)
            y[k] = 1 # 构造 one-hot 的y
            fp = open(file_path,'r')
            x = Image.open(fp)
            x=np.array(x.resize((32,32))) 
            fp.close()
            self.xs.append(x)
            self.ys.append(y)
        #
        self.xs = np.array(self.xs)
        self.ys = np.array(self.ys)
        self.num_examples = len(self.xs)  # 样本数量
        self.curr_pointer = 0
    def next_batch(self, batch_size):
        ''' 随机读取一个批次数据. '''
        start = random.randint(0,self.num_examples-batch_size-1)
        end = start + batch_size
        xx = self.xs[start : end]
        yy = self.ys[start : end]
        #print(xx.shape)
        #print(xx)
        #xx = xx.reshape([1, -1])
        #print(xx.shape)
        rtn_samp = (xx,yy)
        #
        #exit(0)
        #print("-----------------------------1,%d" % batch_size)
        #print(rtn_samp)
        return rtn_samp