Skip to main content

ClikaRT::ops::max_pool

function

max_pool()​

Tensor max_pool(
    Tensor x,
    ClikaRT::Span<const std::int64_t> kernel_size,
    ClikaRT::Span<const std::int64_t> stride,
    ClikaRT::Span<const std::int64_t> padding,
    ClikaRT::Span<const std::int64_t> dilation,
    bool ceil_mode
)

Rank-generic max pooling, channels-last.

x is [N, D1..Dn, C]; the window rank is read from kernel_size's length. Each output element is the maximum over its window. Padding regions never win (they are -inf for the comparison).

Parameters

  • x: the input, [N, D1..Dn, C].
  • kernel_size: window extent per spatial dim.
  • stride: window step per spatial dim; empty = kernel_size.
  • padding: implicit padding per spatial dim.
  • dilation: element spacing inside the window.
  • ceil_mode: round output extents up instead of down; default false.

Returns: the pooled tensor, [N, D1'..Dn', C], dtype of x.

Throws

  • ClikaRT::Error: (INVALID_ARGUMENT) when the parameter ranks disagree or a window over-runs the padded input.

Declared in ClikaRT/compute/ops.h, line 3942