nnf_avg_pool3d function

Avg_pool3d

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 $\lfloor \frac{ \mbox{input planes} }{sT} \rfloor$.

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