张量流量权重矩阵秩误差

Question

import tensorflow as tf
import numpy as np
import os
from PIL import Image
cur_dir = os.getcwd()

def modify_image(image):
   resized = tf.image.resize_images(image, 180, 180, 1)
   resized.set_shape([180,180,3])
   flipped_images = tf.image.flip_up_down(resized)
   return flipped_images

def read_image(filename_queue):
   reader = tf.WholeFileReader()
   key,value = reader.read(filename_queue)
   image = tf.image.decode_jpeg(value)
   return key,image

def inputs():
   filenames = ['standard_1.jpg', 'standard_2.jpg' ]
   filename_queue = tf.train.string_input_producer(filenames)
   filename,read_input = read_image(filename_queue)
   reshaped_image = modify_image(read_input)
   reshaped_image = tf.cast(reshaped_image, tf.float32)
   label=tf.constant([1])
   return reshaped_image,label

def weight_variable(shape):
 initial = tf.truncated_normal(shape, stddev=0.1)
 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')

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



image,label = inputs()
W_conv1=weight_variable([5,5,3,32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)


W_conv2=weights_variable([5,5,32,64])
b_conv2 = bias_variable([32])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
W_fc1 = weight_variable([8 * 8 * 32, 512])
b_fc1 = bias_variable([512])

h_pool2_flat = tf.reshape(h_pool2, [-1, 8*8*32])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
W_fc2 = weight_variable([512, 10])
b_fc2 = bias_variable([10])

y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)

cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))   
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 

init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
tf.train.start_queue_runners(sess=sess)
for i in xrange(100):
    img,label = sess.run(image)
    print (label)
    train_step.run({img, label, 0.5})

当我运行代码时，我收到错误，

"ValueError: ShapesTensorShape([Dimension(180),Dimension(180),Dimension(3)]) and TensorShape([Dimension(None), Dimension(None), Dimension(None), Dimension(None)]) must have the same rank"

但权重已经初始化，即便如此，它也显示为空张量。 正在读取和传输文件和标签。 第一个卷积层有一个5x5窗口，深度为3，我希望有32个这样的5X5滤镜。 因此W_conv1的形状为[5,5,3,32]。

Answer 1

inputs()函数返回180 x 180 x 3的张量，但是tf.nn.conv2d()需要4-D张量的形状batch_size x height x width x num_channels 。

正如etarion建议的那样 ，你可以通过重塑image张量来完成这项工作（例如使用image = tf.expand_dims(image, 0) ）。 但是，如果您正在训练神经网络，您可能希望批量输入。 一种方法是使用tf.train.batch() ：

image, label = inputs()

# Set batch_size to the largest value that works for your configuration.
image_batch, label_batch = tf.train.batch([image, label], batch_size=32)

...然后使用image_batch或label_batch你使用过image和label分别。

Answer 2

看起来输入返回3d张量而conv2d需要4d（第一个维度是批量idx） - 如果你只想运行一个图像，你首先将它重塑为[1,180,180,3]。