lp_pool2d

→Tensor

lp_pool2d(x: Tensor, norm_type: float, kernel_size: int | tuple[int, int], stride: int | tuple[int, int] | None = None, ceil_mode: bool = False)

source

2-D Lp-norm pooling — $\big(\sum |x|^p\big)^{1/p}$ .

Two-dimensional version of lp_pool1d. The exponent $p$ interpolates between sum / average pooling (small $p$ ) and max pooling (large $p$ ), giving the network a tunable "softness" parameter.

Parameters

xTensor

Input of shape (N, C, H, W).

norm_typefloat

Exponent

p > 0

kernel_sizeint or (int, int)

Size of the pooling window.

strideint or (int, int)= None

Window step. Defaults to kernel_size.

ceil_modebool= False

Use ceil instead of floor in the output-size formula.

Returns

Tensor

Output of shape (N, C, H_out, W_out).

Notes

Math (per window $R$ ):

y_{i,c,h,w} = \left( \sum_{(m, n) \in R} |x_{i,c,m,n}|^p \right)^{1/p}

The operator is fully differentiable for any $p > 0$ .

Examples

>>> import lucid
>>> from lucid.nn.functional import lp_pool2d
>>> x = lucid.randn(1, 8, 16, 16)
>>> y = lp_pool2d(x, norm_type=3.0, kernel_size=2)
>>> y.shape
(1, 8, 8, 8)