Question

我是CNN的新手，我正在努力让CNN对手写英文字母（az），（AZ）和数字（0-9）的图像数据进行分类，这些数据集有62个标签。每张图片大小为30 * 30像素。我按照教程https://www.tensorflow.org/get_started/mnist/pros中的步骤操作。当我运行模型时出现错误

tensorflow.python.framework.errors.InvalidArgumentError：不兼容的形状：[40]与[10]

我的批量大小为10，错误似乎在correct_prediction。在Tensorflow Incompatable Shapes Error in Tutorial中找到的同一问题的解决方案并没有解决我的问题。任何帮助将不胜感激。数据集首先在pickle中压缩，这是我的代码。

import tensorflow as tf
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
X = []
y = []
import pickle

#load data from pickle
pickle_file = 'let.pickle'

with open(pickle_file, 'rb') as f:
    save = pickle.load(f)
    X = save['dataset']
    y = save['labels']

    del save  # hint to help gc free up memory

#normalise the features
X = (X - 255/2) / 255

# one hot encoding
y = pd.get_dummies(y)
y = y.values  # change to ndarray
y = np.float32(y)
X = np.float32(X)
Xtr, Xte, Ytr, Yte = train_test_split(X, y, train_size=0.7)

batch_size = 10

sess = tf.InteractiveSession()

x = tf.placeholder(tf.float32, shape=[None, 900])
y_ = tf.placeholder(tf.float32, shape=[None, 62])

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')

W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])

x_image = tf.reshape(x, [-1,30,30,1])

h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)

W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])

h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)

W_fc1 = weight_variable([4 * 4 * 64, 1024])
b_fc1 = bias_variable([1024])

h_pool2_flat = tf.reshape(h_pool2, [-1, 4*4*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

W_fc2 = weight_variable([1024, 62])
b_fc2 = bias_variable([62])

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

cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=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))
sess.run(tf.global_variables_initializer())
for i in range(20000):
    offset = (i * batch_size) % (Ytr.shape[0] - batch_size)
    batch_x = Xtr[offset:(offset + batch_size), :]
    batch_y = Ytr[offset:(offset + batch_size), :]
    if i%100 == 0:
        train_accuracy = accuracy.eval(feed_dict={x:batch_x , y_: batch_y, keep_prob: 1.0})
        print("step %d, training accuracy %g"%(i, train_accuracy))
    train_step.run(feed_dict={x: Xtr[offset:(offset + batch_size), :], y_: Ytr[offset:(offset + batch_size), :], keep_prob: 0.5})

print("test accuracy %g"%accuracy.eval(feed_dict={x: Xte, y_: Yte, keep_prob: 1.0}))

Answer 1

看看这些代码行：

cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))

如果标签是具有正确预测索引的列表（例如：[1,32,13, ...]），则softmax_cross_entropy_with_logits函数是正确的。这意味着错误出现在这些行中。我发表评论说：

tf.argmax(y_conv,1) # Takes the max index of logits
tf.argmax(y_,1) # Takes the max index of ???.

虽然我没有测试它，但用这行代替它应该有效：

correct_prediction = tf.equal(tf.argmax(y_conv,1), y_)

当你修好它时，请告诉我：D

Answer 2

我将完全连接层的输入尺寸从4更改为8，现在正在工作

W_fc1 = weight_variable([8 * 8 * 64, 1024])
b_fc1 = bias_variable([1024])

#the input should be shaped/flattened
h_pool2_flat = tf.reshape(h_pool2, [-1, 8*8*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

张量流上的形状不相容

2 个答案: