Tensorflow收敛但预测不好

我前几天发布了一个类似的问题here，但我已经编辑了我发现的错误，并且预测错误的问题仍然存在 .

我有两个网络 - 一个有3个转换层，另一个有3个转换层，后面跟着3个deconv层 . 两者都采用200x200输入图像 . 输出具有相同的分辨率200x200，但它有两个分类（1的零 - 它是分段网络），因此网络预测维度为200x200x2（加上batch_size） . 我们来谈谈带有deconv层的网络 .

这是奇怪的事情......在10次训练中，其中3次会合并 . 其他7将分解为精度0.0 .

conv和deconv层由ReLu激活 . 优化器做了一些奇怪的事情 . 当我在每次训练迭代后打印预测时，值的大小开始变大 - 这是正确的，考虑到它们都通过ReLu - 但在每次迭代之后，值变小，直到它们大致在0到2之间 . 我随后将它们传递给sigmoid函数（ sigmoid_cross_entropy_wight_logits ） - 从而将大的负值压缩为0，将大的正值压缩为1.当我进行预测时，我通过再次通过sigmoid函数重新激活输出 .

所以在第一次迭代之后，预测值是合理的......

Accuracy = 0.508033
[[[[ 1.  0.]
   [ 0.  1.]
   [ 0.  0.]
   ..., 
   [ 1.  0.]
   [ 1.  1.]
   [ 1.  0.]]

  [[ 0.  1.]
   [ 1.  1.]
   [ 0.  0.]
   ..., 
   [ 1.  1.]
   [ 1.  1.]
   [ 0.  1.]]

但是经过一些迭代后，让我们说它实际上收敛了这个时间，预测值看起来像......（因为优化器使输出更小，它们都在sigmoid函数的奇怪中间地带）

[[ 0.51028508  0.63202268]
   [ 0.24386917  0.52015287]
   [ 0.62086064  0.6953823 ]
   ..., 
   [ 0.2593964   0.13163178]
   [ 0.24617286  0.5210492 ]
   [ 0.24692698  0.5876413 ]]]]
Accuracy = 0.999913

我有错误的优化功能吗？

这是整个代码...跳转到 def conv_net 以查看网络创建...之后该功能是成本函数，优化程序和准确性的定义 . 您会注意到，当我测量准确度并进行预测时，我会使用 tf.nn.sigmoid(pred) 重新激活输出 - 这是因为成本函数 sigmoid_cross_entropy_with_logits 将激活和损失结合在同一函数中 . 换句话说， pred （网络）输出线性值 .

import tensorflow as tf
import pdb
import numpy as np
from numpy import genfromtxt
from PIL import Image

# Parameters
learning_rate = 0.001
training_iters = 10000
batch_size = 10
display_step = 1

# Network Parameters
n_input = 200 # MNIST data input (img shape: 28*28)
n_output = 40000
n_classes = 2 # MNIST total classes (0-9 digits)
#n_input = 200

dropout = 0.75 # Dropout, probability to keep units

# tf Graph input
x = tf.placeholder(tf.float32, [None, n_input, n_input])
y = tf.placeholder(tf.float32, [None, n_input, n_input, n_classes])
keep_prob = tf.placeholder(tf.float32) #dropout (keep probability)


def convert_to_2_channel(x, batch_size):
    #assume input has dimension (batch_size,x,y)
    #output will have dimension (batch_size,x,y,2)
    output = np.empty((batch_size, 200, 200, 2))

    temp_arr1 = np.empty((batch_size, 200, 200))
    temp_arr2 = np.empty((batch_size, 200, 200))

    for i in xrange(batch_size):
        for j in xrange(3):
            for k in xrange(3):
                if x[i][j][k] == 1:
                    temp_arr1[i][j][k] = 1
                    temp_arr2[i][j][k] = 0
                else:
                    temp_arr1[i][j][k] = 0
                    temp_arr2[i][j][k] = 1

    for i in xrange(batch_size):
        for j in xrange(200):
            for k in xrange(200):
                for l in xrange(2):
                    if l == 0:
                        output[i][j][k][l] = temp_arr1[i][j][k]
                    else:
                        output[i][j][k][l] = temp_arr2[i][j][k]

    return output


# Create some wrappers for simplicity
def conv2d(x, W, b, strides=1):
    # Conv2D wrapper, with bias and relu activation
    x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME')
    x = tf.nn.bias_add(x, b)
    return tf.nn.relu(x)

def maxpool2d(x, k=2):
    # MaxPool2D wrapper
    return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
                          padding='SAME')


# Create model
def conv_net(x, weights, biases, dropout):
    # Reshape input picture
    x = tf.reshape(x, shape=[-1, 200, 200, 1])

    # Convolution Layer
    conv1 = conv2d(x, weights['wc1'], biases['bc1'])
    # Max Pooling (down-sampling)
    #conv1 = tf.nn.local_response_normalization(conv1)
    conv1 = maxpool2d(conv1, k=2)

    # Convolution Layer
    conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
    # Max Pooling (down-sampling)
    #conv2 = tf.nn.local_response_normalization(conv2)
    conv2 = maxpool2d(conv2, k=2)

    # Convolution Layer
    conv3 = conv2d(conv2, weights['wc3'], biases['bc3'])
    # # Max Pooling (down-sampling)
    #conv3 = tf.nn.local_response_normalization(conv3)
    conv3 = maxpool2d(conv3, k=2)

    temp_batch_size = tf.shape(x)[0]
    output_shape = [temp_batch_size, 50, 50, 64]
    conv4 = tf.nn.conv2d_transpose(conv3, weights['wdc1'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
    conv4 = tf.nn.bias_add(conv4, biases['bdc1'])
    conv4 = tf.nn.relu(conv4)
    # conv4 = tf.nn.local_response_normalization(conv4)

    # output_shape = tf.pack([temp_batch_size, 100, 100, 32])
    output_shape = [temp_batch_size, 100, 100, 32]
    conv5 = tf.nn.conv2d_transpose(conv4, weights['wdc2'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
    conv5 = tf.nn.bias_add(conv5, biases['bdc2'])
    conv5 = tf.nn.relu(conv5)
    # conv5 = tf.nn.local_response_normalization(conv5)

    # output_shape = tf.pack([temp_batch_size, 200, 200, 1])
    output_shape = [temp_batch_size, 200, 200, 2]
    conv6 = tf.nn.conv2d_transpose(conv5, weights['wdc3'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
    conv6 = tf.nn.bias_add(conv6, biases['bdc3'])
    conv6 = tf.nn.relu(conv6)
    # pdb.set_trace()

    # Fully connected layer
    # Reshape conv2 output to fit fully connected layer input
    fc1 = tf.reshape(conv6, [-1, weights['wd1'].get_shape().as_list()[0]])
    fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
    fc1 = tf.nn.relu(fc1)
    # Apply Dropout
    fc1 = tf.nn.dropout(fc1, dropout)

    return (tf.add(tf.matmul(fc1, weights['out']), biases['out']))# Store layers weight & bias

weights = {
    # 5x5 conv, 1 input, 32 outputs
    'wc1' : tf.Variable(tf.random_normal([5, 5, 1, 32])),
    # 5x5 conv, 32 inputs, 64 outputs
    'wc2' : tf.Variable(tf.random_normal([5, 5, 32, 64])),
    # 5x5 conv, 32 inputs, 64 outputs
    'wc3' : tf.Variable(tf.random_normal([5, 5, 64, 128])),

    'wdc1' : tf.Variable(tf.random_normal([2, 2, 64, 128])),

    'wdc2' : tf.Variable(tf.random_normal([2, 2, 32, 64])),

    'wdc3' : tf.Variable(tf.random_normal([2, 2, 2, 32])),

    # fully connected, 7*7*64 inputs, 1024 outputs
    'wd1': tf.Variable(tf.random_normal([80000, 1024])),
    # 1024 inputs, 10 outputs (class prediction)
    'out': tf.Variable(tf.random_normal([1024, 80000]))
}

biases = {
    'bc1': tf.Variable(tf.random_normal([32])),
    'bc2': tf.Variable(tf.random_normal([64])),
    'bc3': tf.Variable(tf.random_normal([128])),
    'bdc1': tf.Variable(tf.random_normal([64])),
    'bdc2': tf.Variable(tf.random_normal([32])),
    'bdc3': tf.Variable(tf.random_normal([2])),
    'bd1': tf.Variable(tf.random_normal([1024])),
    'out': tf.Variable(tf.random_normal([80000]))
}

# Construct model
pred = conv_net(x, weights, biases, keep_prob)
pred = tf.reshape(pred, [-1,n_input,n_input,n_classes])
# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(pred, y))
# cost = (tf.nn.sigmoid_cross_entropy_with_logits(pred, y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

# Evaluate model
correct_pred = tf.equal(0,tf.cast(tf.sub(tf.nn.sigmoid(pred),y), tf.int32))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver()
# Launch the graph
with tf.Session() as sess:
    sess.run(init)
    summary = tf.train.SummaryWriter('/tmp/logdir/', sess.graph)
    step = 1
    from tensorflow.contrib.learn.python.learn.datasets.scroll import scroll_data
    data = scroll_data.read_data('/home/kendall/Desktop/')
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        batch_x, batch_y = data.train.next_batch(batch_size)
        # Run optimization op (backprop)
        batch_x = batch_x.reshape((batch_size, n_input, n_input))
        batch_y = batch_y.reshape((batch_size, n_input, n_input))
        batch_y = convert_to_2_channel(batch_y, batch_size) #converts the 200x200 ground truth to a 200x200x2 classification
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y,
                                       keep_prob: dropout})
        #measure prediction
        prediction = sess.run(tf.nn.sigmoid(pred), feed_dict={x: batch_x, keep_prob: 1.})
        print prediction
        if step % display_step == 0:
            # Calculate batch loss and accuracdef conv_net(x, weights, biases, dropout):
            save_path = "model.ckpt"
            saver.save(sess, save_path)
            loss, acc = sess.run([cost, accuracy], feed_dict={x: batch_x,
                                                              y: batch_y,
                                                              keep_prob: dropout})
            print "Accuracy = " + str(acc)
            if acc > 0.73:
                break
        step += 1
    print "Optimization Finished!"

    #make prediction
    im = Image.open('/home/kendall/Desktop/HA900_frames/frame0035.tif')
    batch_x = np.array(im)
    # pdb.set_trace()
    batch_x = batch_x.reshape((1, n_input, n_input))
    batch_x = batch_x.astype(float)
    pdb.set_trace()
    prediction = sess.run(tf.nn.sigmoid(pred), feed_dict={x: batch_x, keep_prob: dropout})
    print prediction
    arr1 = np.empty((n_input,n_input))
    arr2 = np.empty((n_input,n_input))
    for i in xrange(n_input):
        for j in xrange(n_input):
            for k in xrange(2):
                if k == 0:
                    arr1[i][j] = (prediction[0][i][j][k])
                else:
                    arr2[i][j] = (prediction[0][i][j][k])
    # prediction = np.asarray(prediction)
    # prediction = np.reshape(prediction, (200,200))
    # np.savetxt("prediction.csv", prediction, delimiter=",")
    np.savetxt("prediction1.csv", arr1, delimiter=",")
    np.savetxt("prediction2.csv", arr2, delimiter=",")
    # np.savetxt("prediction2.csv", arr2, delimiter=",")

    # Calculate accuracy for 256 mnist test images
    print "Testing Accuracy:", \
        sess.run(accuracy, feed_dict={x: data.test.images[:256],
                                      y: data.test.labels[:256],
                                      keep_prob: 1.})

correct_pred 变量（测量精度的变量）是预测和地面实况之间的简单减法运算符，然后与零进行比较（如果两者相等，则差值应为零） .

此外，我已经绘制了网络图，它看起来非常适合我 . 这是一张照片，我不得不裁剪以供查看 .

EDIT: 我发现为什么我的图表看起来很糟糕（感谢Olivier），我也尝试改变我的损失功能，但是没有尽头 - 它仍然在同一个庄园中分道扬..

with tf.name_scope("loss") as scope:
    # cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(pred, y))
    temp_pred = tf.reshape(pred, [-1, 2])
    temp_y = tf.reshape(y, [-1, 2])
    cost = (tf.nn.softmax_cross_entropy_with_logits(temp_pred, temp_y))

EDIT 完整代码现在看起来像这样（仍然发散）

import tensorflow as tf
import pdb
import numpy as np
from numpy import genfromtxt
from PIL import Image

# Parameters
learning_rate = 0.001
training_iters = 10000
batch_size = 10
display_step = 1

# Network Parameters
n_input = 200 # MNIST data input (img shape: 28*28)
n_output = 40000
n_classes = 2 # MNIST total classes (0-9 digits)
#n_input = 200

dropout = 0.75 # Dropout, probability to keep units

# tf Graph input
x = tf.placeholder(tf.float32, [None, n_input, n_input])
y = tf.placeholder(tf.float32, [None, n_input, n_input, n_classes])
keep_prob = tf.placeholder(tf.float32) #dropout (keep probability)


def convert_to_2_channel(x, batch_size):
    #assume input has dimension (batch_size,x,y)
    #output will have dimension (batch_size,x,y,2)
    output = np.empty((batch_size, 200, 200, 2))

    temp_arr1 = np.empty((batch_size, 200, 200))
    temp_arr2 = np.empty((batch_size, 200, 200))

    for i in xrange(batch_size):
        for j in xrange(3):
            for k in xrange(3):
                if x[i][j][k] == 1:
                    temp_arr1[i][j][k] = 1
                    temp_arr2[i][j][k] = 0
                else:
                    temp_arr1[i][j][k] = 0
                    temp_arr2[i][j][k] = 1

    for i in xrange(batch_size):
        for j in xrange(200):
            for k in xrange(200):
                for l in xrange(2):
                    if l == 0:
                        output[i][j][k][l] = temp_arr1[i][j][k]
                    else:
                        output[i][j][k][l] = temp_arr2[i][j][k]

    return output


# Create some wrappers for simplicity
def conv2d(x, W, b, strides=1):
    # Conv2D wrapper, with bias and relu activation
    x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME')
    x = tf.nn.bias_add(x, b)
    return tf.nn.relu(x)

def maxpool2d(x, k=2):
    # MaxPool2D wrapper
    return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
                          padding='SAME')


# Create model
def conv_net(x, weights, biases, dropout):
    # Reshape input picture
    x = tf.reshape(x, shape=[-1, 200, 200, 1])

    with tf.name_scope("conv1") as scope:
    # Convolution Layer
        conv1 = conv2d(x, weights['wc1'], biases['bc1'])
        # Max Pooling (down-sampling)
        #conv1 = tf.nn.local_response_normalization(conv1)
        conv1 = maxpool2d(conv1, k=2)

    # Convolution Layer
    with tf.name_scope("conv2") as scope:
        conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
        # Max Pooling (down-sampling)
        #conv2 = tf.nn.local_response_normalization(conv2)
        conv2 = maxpool2d(conv2, k=2)

    # Convolution Layer
    with tf.name_scope("conv3") as scope:
        conv3 = conv2d(conv2, weights['wc3'], biases['bc3'])
        # # Max Pooling (down-sampling)
        #conv3 = tf.nn.local_response_normalization(conv3)
        conv3 = maxpool2d(conv3, k=2)


    temp_batch_size = tf.shape(x)[0]
    with tf.name_scope("deconv1") as scope:
        output_shape = [temp_batch_size, 50, 50, 64]
        conv4 = tf.nn.conv2d_transpose(conv3, weights['wdc1'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
        conv4 = tf.nn.bias_add(conv4, biases['bdc1'])
        conv4 = tf.nn.relu(conv4)
        # conv4 = tf.nn.local_response_normalization(conv4)

    with tf.name_scope("deconv2") as scope:
        # output_shape = tf.pack([temp_batch_size, 100, 100, 32])
        output_shape = [temp_batch_size, 100, 100, 32]
        conv5 = tf.nn.conv2d_transpose(conv4, weights['wdc2'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
        conv5 = tf.nn.bias_add(conv5, biases['bdc2'])
        conv5 = tf.nn.relu(conv5)
        # conv5 = tf.nn.local_response_normalization(conv5)

    with tf.name_scope("deconv3") as scope:
        # output_shape = tf.pack([temp_batch_size, 200, 200, 1])
        output_shape = [temp_batch_size, 200, 200, 2]
        conv6 = tf.nn.conv2d_transpose(conv5, weights['wdc3'], output_shape=output_shape, strides=[1,2,2,1], padding="VALID")
        conv6 = tf.nn.bias_add(conv6, biases['bdc3'])
    # conv6 = tf.nn.relu(conv6)
    # pdb.set_trace()
    conv6 = tf.nn.dropout(conv6, dropout)

    return conv6
    # Fully connected layer
    # Reshape conv2 output to fit fully connected layer input
    # fc1 = tf.reshape(conv6, [-1, weights['wd1'].get_shape().as_list()[0]])
    # fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
    # fc1 = tf.nn.relu(fc1)
    # # Apply Dropout
    # fc1 = tf.nn.dropout(fc1, dropout)
    #
    # return (tf.add(tf.matmul(fc1, weights['out']), biases['out']))# Store layers weight & bias

weights = {
    # 5x5 conv, 1 input, 32 outputs
    'wc1' : tf.Variable(tf.random_normal([5, 5, 1, 32])),
    # 5x5 conv, 32 inputs, 64 outputs
    'wc2' : tf.Variable(tf.random_normal([5, 5, 32, 64])),
    # 5x5 conv, 32 inputs, 64 outputs
    'wc3' : tf.Variable(tf.random_normal([5, 5, 64, 128])),

    'wdc1' : tf.Variable(tf.random_normal([2, 2, 64, 128])),

    'wdc2' : tf.Variable(tf.random_normal([2, 2, 32, 64])),

    'wdc3' : tf.Variable(tf.random_normal([2, 2, 2, 32])),

    # fully connected, 7*7*64 inputs, 1024 outputs
    'wd1': tf.Variable(tf.random_normal([80000, 1024])),
    # 1024 inputs, 10 outputs (class prediction)
    'out': tf.Variable(tf.random_normal([1024, 80000]))
}

biases = {
    'bc1': tf.Variable(tf.random_normal([32])),
    'bc2': tf.Variable(tf.random_normal([64])),
    'bc3': tf.Variable(tf.random_normal([128])),
    'bdc1': tf.Variable(tf.random_normal([64])),
    'bdc2': tf.Variable(tf.random_normal([32])),
    'bdc3': tf.Variable(tf.random_normal([2])),
    'bd1': tf.Variable(tf.random_normal([1024])),
    'out': tf.Variable(tf.random_normal([80000]))
}

# Construct model
# with tf.name_scope("net") as scope:
pred = conv_net(x, weights, biases, keep_prob)
pred = tf.reshape(pred, [-1,n_input,n_input,n_classes])
# Define loss and optimizer
with tf.name_scope("loss") as scope:
    # cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(pred, y))
    temp_pred = tf.reshape(pred, [-1, 2])
    temp_y = tf.reshape(y, [-1, 2])
    cost = (tf.nn.softmax_cross_entropy_with_logits(temp_pred, temp_y))

with tf.name_scope("opt") as scope:
    optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
    # optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(cost)


# Evaluate model
with tf.name_scope("acc") as scope:
    correct_pred = tf.equal(0,tf.cast(tf.sub(tf.nn.softmax(temp_pred),y), tf.int32))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver()
# Launch the graph
with tf.Session() as sess:
    sess.run(init)
    summary = tf.train.SummaryWriter('/tmp/logdir/', sess.graph)
    step = 1
    from tensorflow.contrib.learn.python.learn.datasets.scroll import scroll_data
    data = scroll_data.read_data('/home/kendall/Desktop/')
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        batch_x, batch_y = data.train.next_batch(batch_size)
        # Run optimization op (backprop)
        batch_x = batch_x.reshape((batch_size, n_input, n_input))
        batch_y = batch_y.reshape((batch_size, n_input, n_input))
        batch_y = convert_to_2_channel(batch_y, batch_size) #converts the 200x200 ground truth to a 200x200x2 classification
        batch_y = batch_y.reshape(batch_size * n_input * n_input, 2)
        sess.run(optimizer, feed_dict={x: batch_x, temp_y: batch_y,
                                       keep_prob: dropout})
        #measure prediction
        prediction = sess.run(tf.nn.softmax(temp_pred), feed_dict={x: batch_x, keep_prob: dropout})
        print prediction
        if step % display_step == 0:
            # Calculate batch loss and accuracdef conv_net(x, weights, biases, dropout):
            save_path = "model.ckpt"
            saver.save(sess, save_path)
            loss, acc = sess.run([cost, accuracy], feed_dict={x: batch_x,
                                                              y: batch_y,
                                                              keep_prob: dropout})
            print "Accuracy = " + str(acc)
            if acc > 0.73:
                break
        step += 1
    print "Optimization Finished!"

    #make prediction
    im = Image.open('/home/kendall/Desktop/HA900_frames/frame0035.tif')
    batch_x = np.array(im)
    # pdb.set_trace()
    batch_x = batch_x.reshape((1, n_input, n_input))
    batch_x = batch_x.astype(float)
    pdb.set_trace()
    prediction = sess.run(tf.nn.sigmoid(pred), feed_dict={x: batch_x, keep_prob: dropout})
    print prediction
    arr1 = np.empty((n_input,n_input))
    arr2 = np.empty((n_input,n_input))
    for i in xrange(n_input):
        for j in xrange(n_input):
            for k in xrange(2):
                if k == 0:
                    arr1[i][j] = (prediction[0][i][j][k])
                else:
                    arr2[i][j] = (prediction[0][i][j][k])
    # prediction = np.asarray(prediction)
    # prediction = np.reshape(prediction, (200,200))
    # np.savetxt("prediction.csv", prediction, delimiter=",")
    np.savetxt("prediction1.csv", arr1, delimiter=",")
    np.savetxt("prediction2.csv", arr2, delimiter=",")
    # np.savetxt("prediction2.csv", arr2, delimiter=",")

    # Calculate accuracy for 256 mnist test images
    print "Testing Accuracy:", \
        sess.run(accuracy, feed_dict={x: data.test.images[:256],
                                      y: data.test.labels[:256],
                                      keep_prob: 1.})