What is the essential difference between `tf.contrib.layers.max_pool2d` and `tf.nn.max_pool`?

Question

I found two functions, tf.contrib.layers.max_pool2d and tf.nn.max_pool, for max pooling in TensorFlow.

tf.contrib.layers.max_pool2d seems to be a wrapper of tf.nn.max_pool, but I could not understand what is the advantage of using `tf.contrib.layers.max_pool2d.

What is the essential difference between tf.contrib.layers.max_pool2d and tf.nn.max_pool?

Resources:

• document of tf.contrib.layers.max_pool2d
  • https://www.tensorflow.org/versions/master/api_docs/python/contrib.layers.html#max_pool2d
• document of tf.nn.max_pool
  • https://www.tensorflow.org/versions/master/api_docs/python/nn.html#max_pool

Answer 1

Well, functionally, there is no "essential" difference between them, except for some minor differences between their signatures(ksize, strides, etc.)

But please pay attention, tf.contrib is an unstable collection of some high-level functions contributed from various sources, thus might be modified or even removed in future versions.