tf.pad returns 数组用零填充，MNIST 数据集从 (28,28) 图像大小填充到 (32,32)

Question

我需要在 MNIST 数据集上训练 LeNet5 CNN。关于 LeNet5 的原始论文，我想在输入图像大小 32 x 32 而不是图像大小 28 x 28 上进行训练，以具有相同数量的参数。

我尝试使用 tf.pad 和 numpy.pad 函数，但我遇到了同样的问题：我可以使用 (31,32) 或 (32,31) returned图像大小，但当我尝试 (32,32) 或更多时，函数 return 数组填充零而不是 X_train 原始值 + 填充值。这里的代码和结果：

(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()

print(X_train.shape)
print('Before padding',X_train[0])

X_train_pad = tf.pad(X_train,([0,0],[2,2],[2,2]), mode='CONSTANT', constant_values=0, name=None)

print('After padding',X_train_pad[0])

(60000, 28, 28)

Before padding [[  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   3  18  18  18 126 136
  175  26 166 255 247 127   0   0   0   0]
 [  0   0   0   0   0   0   0   0  30  36  94 154 170 253 253 253 253 253
  225 172 253 242 195  64   0   0   0   0]
 [  0   0   0   0   0   0   0  49 238 253 253 253 253 253 253 253 253 251
   93  82  82  56  39   0   0   0   0   0]
 [  0   0   0   0   0   0   0  18 219 253 253 253 253 253 198 182 247 241
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0  80 156 107 253 253 205  11   0  43 154
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0  14   1 154 253  90   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0 139 253 190   2   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0  11 190 253  70   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0  35 241 225 160 108   1
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0  81 240 253 253 119
   25   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0  45 186 253 253
  150  27   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0  16  93 252
  253 187   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 249
  253 249  64   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0  46 130 183 253
  253 207   2   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0  39 148 229 253 253 253
  250 182   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0  24 114 221 253 253 253 253 201
   78   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0  23  66 213 253 253 253 253 198  81   2
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0  18 171 219 253 253 253 253 195  80   9   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0  55 172 226 253 253 253 253 244 133  11   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0 136 253 253 253 212 135 132  16   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]
 [  0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0
    0   0   0   0   0   0   0   0   0   0]]

After padding tf.Tensor(
[[0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 ...
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]
 [0 0 0 ... 0 0 0]], shape=(32, 32), dtype=uint8)

当我用 1 填充时，我可以看到填充值和 0 而不是 X_train 的原始值：

X_train_pad = tf.pad(X_train,([0,0],[2,2],[2,2]), mode='CONSTANT', constant_values=1, name=None)

After padding tf.Tensor(
[[1 1 1 ... 1 1 1]
 [1 1 1 ... 1 1 1]
 [1 1 0 ... 0 1 1]
 ...
 [1 1 0 ... 0 1 1]
 [1 1 1 ... 1 1 1]
 [1 1 1 ... 1 1 1]], shape=(32, 32), dtype=uint8)

你知道我的错误是什么吗？

谢谢

Answer 1

没有错

填充工作正常。

您正在目视检查它并得出它不正确的结论，而不是使用您的计算机来简单地测试它。

import numpy as np
import tensorflow as tf

# get data
(X_train, _), (_, _) = tf.keras.datasets.mnist.load_data()

# pad
X_train_pad = tf.pad(X_train, [[0, 0], [2, 2], [2, 2]], mode="CONSTANT")

# extract the original image by manually removing the
# padding from the borders
extract_original_image = X_train_pad[:, 2:-2, 2:-2]

print(extract_original_image.shape)
# TensorShape([60000, 28, 28])

# check padded image with padding removed equals original
np.testing.assert_equal(extract_original_image.numpy(), X_train)

# check padding is all zeros
assert tf.math.reduce_sum(X_train_pad[:, :2, :2]).numpy() == 0
assert tf.math.reduce_sum(X_train_pad[:, -2:, -2:]).numpy() == 0

# check the sums are the same
assert tf.math.reduce_sum(X_train_pad) == tf.math.reduce_sum(X_train)

tf.pad returns 数组用零填充，MNIST 数据集从 (28,28) 图像大小填充到 (32,32)

tf.pad returns array filled with zero, MNIST dataset padding from (28,28) image size to (32,32)

python

numpy

padding

mnist

tensorflow