mnist整合mobilenet

#coding=utf8#author by fffupeng#envirnment anaconda python 3.5 windows#这是一个将mnist中的卷积改成mobilenet的思想import tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets('MNIST_data',one_hot=True)def weight_variable(shape):    initial = tf.truncated_normal(shape,stddev=0.1)    #从截断的正态分布中输出随机值。    #生成的值服从具有指定平均值和标准偏差的正态分布，如果生成的值大于平均值2个标准偏差的值则丢弃重新选择。    return tf.Variable(initial)def bias_variable(shape):    initial = tf.constant(0.1,shape=shape)    return tf.Variable(initial)def conv2d(x,W):    return tf.nn.conv2d(x,W,strides = [1,1,1,1],padding='SAME') #http://www.jianshu.com/p/05c4f1621c7e                                                                #VALID不会添加新元素，SAME添加元素使得卷积前后尺寸不变def max_pool_2x2(x):    return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')x = tf.placeholder(tf.float32,[None,784])y_actual = tf.placeholder(tf.float32,[None,10])x_image = tf.reshape(x,[-1,28,28,1])    ##conv1 输入 [-1,28,28,1]输出[-1,28,28,32]#注意这里没有考虑池化，BN,relu，仅仅是将卷积分解#分解之后的卷积应该是[-1,28,28,1]--->[-1,28,28,1]---->[-1,28,28,32]# w1 = weight_variable([5,5,1,32])# b1 = bias_variable([32])# conv1 = tf.nn.relu(conv2d(x_image,w1)+b1)conv1_w1 = weight_variable([5,5,1,1])conv1_b1 = weight_variable([1])conv1_y1 = conv2d(x_image,conv1_w1)+conv1_b1conv1_w2 = weight_variable([1,1,1,32])conv1_b2 = weight_variable([32])conv1_y2 = conv2d(conv1_y1,conv1_w2)+conv1_b2pool1 = max_pool_2x2(conv1_y2)#输出-1x14x14x32#第二个卷积层# w2 = weight_variable([5,5,32,64])   # 5x5的卷积，32维度的输入， 64 维度的输出# b2 = bias_variable([64])# conv2 = tf.nn.relu(conv2d(pool1,w2)+b2)conv2_w1 = weight_variable([5,5,32,32])conv2_b1 = weight_variable([32])conv2_y1 = conv2d(pool1,conv2_w1)+conv2_b1conv2_w2 = weight_variable([1,1,32,64])conv2_b2 = weight_variable([64])conv2_y2 = conv2d(conv2_y1,conv2_w2)+conv2_b2pool2 = max_pool_2x2(conv2_y2) #输出7x7x64#全连接层1fc_w1 = weight_variable([7 * 7 * 64, 1024])fc_b1 = bias_variable([1024])#将pool2特征层展平pool2_flat = tf.reshape(pool2,[-1,7*7*64])fc1 =tf.nn.relu(tf.matmul(pool2_flat,fc_w1)+fc_b1)#全连接层2fc_w2 = weight_variable([1024,10])fc_b2 = weight_variable([10])y = tf.matmul(fc1,fc_w2)+fc_b2loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_actual,logits=y))train_step = tf.train.AdamOptimizer(0.0001).minimize(loss)correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_actual, 1))accuracy = tf.reduce_sum(tf.cast(correct_prediction, tf.float32))   #计算正确的个数def mytrain():    saver = tf.train.Saver()    with tf.Session() as sess:            sess.run(tf.global_variables_initializer())            for i in range(1000):                batch_x, batch_y = mnist.train.next_batch(1000)                if i%10 == 0:                    print('train acc: ' +str(sess.run(accuracy,feed_dict={x: batch_x, y_actual: batch_y})/1000)+'%')                sess.run(train_step,feed_dict={x:batch_x,y_actual:batch_y})            saver.save(sess, "D:\workspace\save_model\mnist-mobilenet\mnist.ckpt")#window下面保存的保存路径还是写绝对路径比较安全            #如果显存够可以使用下面的测试            #print(sess.run(accuracy,feed_dict={x: mnist.train.images, y_actual: mnist.train.labels}))            #print(sess.run(accuracy,feed_dict={x:mnist.test.images,y_actual:mnist.test.labels}))#保存读取模型参考 http://blog.csdn.net/marsjhao/article/details/72829635def mytest():    saver = tf.train.Saver()    with tf.Session() as sess:            #saver = tf.train.import_meta_graph("D:\workspace\save_model\mnist3\mnist.ckpt.meta")        saver.restore(sess, r"D:\workspace\save_model\mnist-mobilenet\mnist.ckpt")  #        right = 0        for i in range(10):            test_x,test_y = mnist.test.next_batch(1000)            #注意这里将accuracy计算方式改掉了，之前是直接计算准确率，现在是统计正确的个数            right = right + (sess.run(accuracy,feed_dict={x:test_x,y_actual:test_y}))        print('test acc: '+ str(right/10000))if __name__ == '__main__':    mytrain()   #>>>train acc: 0.973%    mytest()   #>>> test acc: 0.9741