View source on GitHub

Performs m x n pooling through a combination of 1xm and 1xn pooling.

input_tensor Input tensor. Must be rank 2
window_shape Pooling window shape
pooling_type Either 'MAX' or 'AVG'
strides The stride of the pooling window
padding 'SAME' or 'VALID'.
name Name of the op

A rank 2 tensor containing the pooled output

ValueError if the input tensor is not rank 2