Skip to contents

Applies 3D average-pooling operation in \(kT * kH * kW\) regions by step size \(sT * sH * sW\) steps. The number of output features is equal to \(\lfloor \frac{ \mbox{input planes} }{sT} \rfloor\).

Usage

nnf_avg_pool3d(
  input,
  kernel_size,
  stride = NULL,
  padding = 0,
  ceil_mode = FALSE,
  count_include_pad = TRUE,
  divisor_override = NULL
)

Arguments

input

input tensor (minibatch, in_channels , iT * iH , iW)

kernel_size

size of the pooling region. Can be a single number or a tuple (kT, kH, kW)

stride

stride of the pooling operation. Can be a single number or a tuple (sT, sH, sW). Default: kernel_size

padding

implicit zero paddings on both sides of the input. Can be a single number or a tuple (padT, padH, padW), Default: 0

ceil_mode

when True, will use ceil instead of floor in the formula to compute the output shape

count_include_pad

when True, will include the zero-padding in the averaging calculation

divisor_override

NA if specified, it will be used as divisor, otherwise size of the pooling region will be used. Default: NULL