python-3.x - TensorFlow:为什么avg_pool忽略一个步幅维度？

Question

我试图跨越 channel 维度，并且以下代码表现出令人惊讶的行为。我期望tf.nn.max_pool和tf.nn.avg_pool在输入完全相同的参数时应产生相​​同形状的张量。不是这种情况。

import tensorflow as tf

x = tf.get_variable('x', shape=(100, 32, 32, 64),
        initializer=tf.constant_initializer(5), dtype=tf.float32)
ksize = (1, 2, 2, 2)
strides = (1, 2, 2, 2)
max_pool = tf.nn.max_pool(x, ksize, strides, padding='SAME')
avg_pool = tf.nn.avg_pool(x, ksize, strides, padding='SAME')
print(max_pool.shape)
print(avg_pool.shape)

此打印

$ python ex04/mini.py 
(100, 16, 16, 32)
(100, 16, 16, 64)

显然，我误会了一些东西。

Answer 1

链接https://github.com/Hvass-Labs/TensorFlow-Tutorials/issues/19指出:

The first and last stride must always be 1, because the first is for the image-number and the last is for the input-channel.